feat(training): complete training knowledge v1

.agents/skills/trainlog-anatomy/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: trainlog-anatomy
+description: Use Trainlog's cited anatomy and biomechanics knowledge for exercise targeting, BODY ZONES, equipment interpretation, substitution, MAX context, and future workout or program design.
+---
+
+# Trainlog anatomy
+
+Resolve domain decisions from scientific evidence before implementation.
+Start at `docs/domain/` from the repository root and consult the relevant
+canonical assets in `catalog/`:
+
+- `science-references-v1.json`: evidence, bibliographic metadata and limitations;
+- `muscles-v1.json` and `joint-actions-v1.json`: functional anatomy;
+- `movement-patterns-v1.json`: explicitly defined programming abstractions;
+- `exercise-knowledge-v1.json`: variant-specific roles and BODY ZONE projection;
+- `equipment-knowledge-v1.json`: physical equipment and compatible exercises.
+
+Follow `source_refs` to the cited references. Check that evidence supports the
+actual variant and claim; structural validation is not scientific validation.
+Use `anatomie` for substantive new research and `anatomie-xhigh` only when one
+substantive investigation leaves material scientific ambiguity unresolved.
+Scientific conclusions must be settled before implementation agents encode them.
+
+Preserve these distinctions:
+
+- Physical machine != exercise != movement pattern != joint action != muscle.
+- BODY ZONE is a business/UX projection, not a complete anatomical taxonomy.
+- Higher EMG amplitude does not establish better hypertrophy, strength outcomes,
+  or universal primary-muscle status.
+- Manufacturer documentation identifies mechanics; it does not establish
+  anatomical outcomes. Do not invent a manufacturer, model or trajectory.
+
+Separate established anatomy, biomechanical interpretation, EMG, intervention
+evidence, manufacturer statements and practical inference. Retain `high`,
+`moderate` or `uncertain` confidence; downgrade unsupported claims. Unknown
+custom exercises stay unclassified until their actual execution is established.
+
+Use existing exercise/equipment identities. Never infer runtime knowledge from
+names, merge multifunction-machine exercises, overwrite persisted BODY ZONES,
+or copy user history into scientific catalogs. MAX belongs to an exercise and
+its equipment/resistance/execution context; assistance is not external load.
+
+TRAINING KNOWLEDGE V1 provides read-only knowledge and history composition.
+Workout/program prescription and clinical rehabilitation are outside its scope.

AGENTS.md

@@ -324,3 +324,13 @@ A task is complete only when:
 - synchronization remains idempotent where applicable;
 - documentation describes the resulting state;
 - no known regression is intentionally left behind.
+
+## 14. Training knowledge
+
+Before domain decisions involving anatomy, biomechanics, exercise targeting,
+BODY ZONES, machine/exercise interpretation, substitution, MAX interpretation,
+workout generation or program generation, consult the project
+[`trainlog-anatomy` skill](.agents/skills/trainlog-anatomy/SKILL.md),
+`docs/domain/`, the scientific catalogs in `catalog/`, and their cited references.
+Scientific knowledge is separate from runtime user data; preserve explicit
+uncertainty and settle scientific conclusions before encoding behavior.

CHANGELOG.md

@@ -9,6 +9,19 @@ Detailed implementation chronology remains available in Git history and
 
 ### Added
 
+- `TRAINING_KNOWLEDGE_V1` read-only scientific knowledge infrastructure:
+  six authored, versioned JSON catalogs with cited references; deterministic C
+  generation; Android immutable asset loading; stable-ID catalog queries; and
+  desktop/Android composition of persisted exercise zones, compatible
+  equipment, explicit MAX and bounded occurrence/set history. The feature
+  introduces no schema migration, database seeding, synchronization artifact,
+  runtime prescription or generated UUID association. Scientific review,
+  independent temporal review, final engineering review, repair verification,
+  and final executable validation passed. The initial audit's four findings
+  were closed by one bounded repair chain. See
+  `docs/reviews/training_knowledge_v1_temporal_contract.md` and
+  `docs/domain/knowledge_system.md`.
+
 - `BODY_ZONES_V1`: the canonical `catalog/body-zones-v1.json` taxonomy with
   stable IDs, French display metadata, hierarchy, deterministic sort order and
   exact stable-exercise-ID migration evidence;

README.md

@@ -42,10 +42,25 @@ BODY_ZONE_SYNC_V1=PASS
 BODY_ZONES_DESKTOP_REAL_MIGRATION=PASS
 BODY_ZONES_ANDROID_DEVICE_VALIDATION=PASS
 
-DESKTOP_TESTS=39/39 PASS
+DESKTOP_TESTS=42/42 PASS (recorded validation checkpoint)
 ANDROID_BUILD=PASS
+
+TRAINING_KNOWLEDGE_V1=PASS
 ```
 
+`TRAINING_KNOWLEDGE_V1` has passed its bounded scientific review, independent
+temporal delta review, final engineering audit, repair verification, and final
+executable validation. Its C and
+Android occurrence pages and latest-MAX context share the settled exact-instant
+ordering and source-text cursor contract. The independent temporal review
+returned PASS with no findings. The initial full-tranche audit's four findings
+were resolved by one bounded repair chain and independently verified. Fresh
+validation passed: strict build, 42 Meson tests, Python knowledge/temporal
+tests, validators, headers, sanitizers, 12-form temporal probes, Android 56
+tests with one known fixture skip, and Java 17 debug assembly.
+The [temporal correction record](docs/reviews/training_knowledge_v1_temporal_contract.md)
+defines the accepted grammar, regression evidence and remaining review boundary.
+
 ## Architecture
 
 ```text
@@ -255,6 +270,8 @@ Exercise-zone metadata travels separately in the sole bidirectional
 - `docs/exchange_format.md`: frozen Trainlog JSON v1 contract;
 - `docs/tests.md`: validation strategy;
 - `docs/roadmap.md`: completed gates and future cursor;
+- `docs/domain/knowledge_system.md`: read-only scientific knowledge system,
+  runtime-context boundary and documented future planning pipeline;
 - `AGENTS.md`: development contract.
 
 ## Development principles

android/app/src/main/java/com/labfytools/trainlog/data/TrainingKnowledgeCatalog.kt

@@ -0,0 +1,383 @@
+package com.labfytools.trainlog.data
+
+import android.content.Context
+import org.json.JSONArray
+import org.json.JSONObject
+import java.util.Locale
+
+enum class KnowledgeConfidence { HIGH, MODERATE, UNCERTAIN }
+enum class MuscleEntityType { MUSCLE, MUSCLE_REGION, MUSCLE_GROUP }
+enum class ExerciseKnowledgeStatus { RESOLVED_FAMILY_VARIANT_LIMITED, CONDITIONAL, UNRESOLVED }
+enum class BodyZoneAuditStatus { CONFIRMED, QUESTIONABLE, UNRESOLVED }
+enum class BodyZoneAuditSeverity { NONE, INFORMATION }
+enum class EquipmentScienceStatus {
+    SCIENTIFICALLY_DOCUMENTED,
+    MECHANICALLY_IDENTIFIED_ANATOMY_INCOMPLETE,
+    EQUIPMENT_IDENTITY_UNCERTAIN,
+}
+enum class MuscleRole { PRIMARY, SECONDARY, STABILIZERS }
+
+data class ScienceReference(
+    val refId: String, val title: String, val authorsOrOrganization: String, val year: Int?,
+    val type: String, val url: String, val topics: List<String>, val notes: String,
+    val limitations: String, val doi: String?, val pmid: String?, val accessedOn: String,
+    val publicationNote: String?,
+)
+
+data class MuscleKnowledge(
+    val muscleId: String, val displayName: String, val displayNameFr: String,
+    val entityType: MuscleEntityType, val anatomicalGroup: String, val aggregateGroupId: String?,
+    val bodyZoneId: String, val bodyZoneIds: List<String>, val jointActionIds: List<String>,
+    val primaryActions: List<String>, val primaryActionsSemantics: String,
+    val overlapWarning: String, val functionalNotes: String, val confidence: KnowledgeConfidence,
+    val evidenceType: String, val sourceRefs: List<String>, val memberMuscleIds: List<String>,
+)
+
+data class JointActionKnowledge(
+    val actionId: String, val displayNameFr: String, val definition: String,
+    val anatomicalRegion: String, val jointComplex: String, val jointOrComplex: String,
+    val principalPlane: String, val planeNotes: String, val contributingMuscleIds: List<String>,
+    val contributorSemantics: String, val evidenceType: String, val confidence: KnowledgeConfidence,
+    val sourceRefs: List<String>, val notes: String,
+)
+
+data class MovementPatternKnowledge(
+    val patternId: String, val displayNameFr: String, val definition: String,
+    val typicalActionIds: List<String>, val typicalBodyZoneIds: List<String>,
+    val evidenceType: String, val confidence: KnowledgeConfidence, val sourceRefs: List<String>,
+    val notes: String, val bodyZoneSemantics: String?, val parentPatternId: String?,
+)
+
+data class ExerciseInterpretation(
+    val familyDescription: String, val actionIds: List<String>, val patternIds: List<String>,
+    val primaryMuscleIds: List<String>, val secondaryMuscleIds: List<String>,
+    val stabilizerMuscleIds: List<String>, val primaryZoneId: String,
+    val secondaryZoneIds: List<String>, val confidence: KnowledgeConfidence,
+    val evidenceType: String, val sourceRefs: List<String>, val variantNotes: String,
+    val roleNotes: String, val requiredConfirmation: String? = null,
+) {
+    fun muscles(role: MuscleRole): List<String> = when (role) {
+        MuscleRole.PRIMARY -> primaryMuscleIds
+        MuscleRole.SECONDARY -> secondaryMuscleIds
+        MuscleRole.STABILIZERS -> stabilizerMuscleIds
+    }
+}
+
+data class ScientificBodyZoneMapping(val primaryZoneId: String, val secondaryZoneIds: List<String>)
+data class ExistingBodyZoneMapping(val primaryZoneId: String?, val secondaryZoneIds: List<String>)
+
+data class BodyZoneAudit(
+    val status: BodyZoneAuditStatus, val severity: BodyZoneAuditSeverity,
+    val confidence: KnowledgeConfidence, val rationale: String,
+    val existingPrimaryZoneId: String?, val existingSecondaryZoneIds: List<String>,
+    val proposedMutation: String?, val sourceRefs: List<String>,
+)
+
+data class ExerciseKnowledge(
+    val exerciseId: String, val exerciseName: String, val equipmentIds: List<String>,
+    val identityEvidence: String, val identityEvidenceType: String,
+    val resolutionStatus: ExerciseKnowledgeStatus, val confidence: KnowledgeConfidence,
+    val interpretation: ExerciseInterpretation?, val conditionalInterpretation: ExerciseInterpretation?,
+    val existingBodyZones: ExistingBodyZoneMapping, val sourceRefs: List<String>,
+    val limitations: List<String>, val equipmentLinkStatus: String?, val bodyZoneAudit: BodyZoneAudit,
+)
+
+data class EquipmentCapabilityKnowledge(
+    val displayName: String, val exerciseIds: List<String>, val linkStatus: String,
+    val interpretation: ExerciseInterpretation?, val requirements: String,
+)
+
+data class EquipmentKnowledge(
+    val equipmentId: String, val manufacturer: String?, val model: String?,
+    val identificationStatus: String, val scienceStatus: EquipmentScienceStatus,
+    val catalogType: String, val catalogLoadSemantics: EquipmentLoadSemantics?,
+    val mechanics: String, val evidenceType: String, val confidence: KnowledgeConfidence,
+    val sourceRefs: List<String>, val capabilities: List<EquipmentCapabilityKnowledge>,
+    val requiresActualExercise: Boolean, val limitations: List<String>, val auditNote: String?,
+    val scientificStatusScope: String,
+)
+
+data class KnowledgeExerciseFilters(
+    val movementPatternId: String? = null,
+    val muscleId: String? = null,
+    val muscleRole: MuscleRole? = null,
+    val scientificZoneId: String? = null,
+    val includeZoneDescendants: Boolean = true,
+    val equipmentId: String? = null,
+)
+
+/**
+ * Immutable TRAINING KNOWLEDGE V1 view over the six shared repository assets.
+ * WHY: strict loading makes malformed scientific data a deployment failure and
+ * prevents Android from quietly developing a second, hand-maintained taxonomy.
+ */
+class TrainingKnowledgeCatalog private constructor(
+    val references: List<ScienceReference>, val muscles: List<MuscleKnowledge>,
+    val jointActions: List<JointActionKnowledge>, val movementPatterns: List<MovementPatternKnowledge>,
+    val exercises: List<ExerciseKnowledge>, val equipment: List<EquipmentKnowledge>,
+    private val bodyZones: BodyZoneCatalog,
+) {
+    private val referencesById = references.associateBy { it.refId }
+    private val musclesById = muscles.associateBy { it.muscleId }
+    private val actionsById = jointActions.associateBy { it.actionId }
+    private val patternsById = movementPatterns.associateBy { it.patternId }
+    private val exercisesById = exercises.associateBy { it.exerciseId }
+    private val equipmentById = equipment.associateBy { it.equipmentId }
+
+    fun getReference(refId: String) = referencesById[refId]
+    fun getMuscle(muscleId: String) = musclesById[muscleId]
+    fun getJointAction(actionId: String) = actionsById[actionId]
+    fun getMovementPattern(patternId: String) = patternsById[patternId]
+    fun getExerciseKnowledge(exerciseId: String): ExerciseKnowledge? = exercisesById[exerciseId]
+    fun getConditionalExerciseKnowledge(exerciseId: String): ExerciseKnowledge? =
+        exercisesById[exerciseId]?.takeIf { it.resolutionStatus == ExerciseKnowledgeStatus.CONDITIONAL }
+    fun getEquipmentKnowledge(equipmentId: String): EquipmentKnowledge? = equipmentById[equipmentId]
+
+    fun getScientificBodyZoneMapping(exerciseId: String): ScientificBodyZoneMapping? =
+        resolvedInterpretation(exerciseId)?.let { ScientificBodyZoneMapping(it.primaryZoneId, it.secondaryZoneIds) }
+
+    fun listExercisesByMovementPattern(patternId: String): List<ExerciseKnowledge> {
+        require(patternsById.containsKey(patternId)) { "pattern_id inconnu: $patternId" }
+        return queryExercises(KnowledgeExerciseFilters(movementPatternId = patternId))
+    }
+
+    fun listExercisesByMuscle(muscleId: String, role: MuscleRole): List<ExerciseKnowledge> {
+        require(musclesById.containsKey(muscleId)) { "muscle_id inconnu: $muscleId" }
+        return queryExercises(KnowledgeExerciseFilters(muscleId = muscleId, muscleRole = role))
+    }
+
+    fun listCompatibleExercises(equipmentId: String): List<ExerciseKnowledge> {
+        require(equipmentById.containsKey(equipmentId)) { "equipment_id inconnu: $equipmentId" }
+        return queryExercises(KnowledgeExerciseFilters(equipmentId = equipmentId))
+    }
+
+    fun queryExercises(filters: KnowledgeExerciseFilters): List<ExerciseKnowledge> {
+        filters.movementPatternId?.let { require(patternsById.containsKey(it)) { "pattern_id inconnu: $it" } }
+        filters.muscleId?.let { require(musclesById.containsKey(it)) { "muscle_id inconnu: $it" } }
+        require((filters.muscleId == null) == (filters.muscleRole == null)) {
+            "muscleId et muscleRole doivent être fournis ensemble"
+        }
+        filters.equipmentId?.let { require(equipmentById.containsKey(it)) { "equipment_id inconnu: $it" } }
+        val zones = filters.scientificZoneId?.let {
+            require(bodyZones.lookup(it) != null) { "zone_id inconnu: $it" }
+            if (filters.includeZoneDescendants) bodyZones.descendantsAndSelf(it) else setOf(it)
+        }
+        return exercises.filter { exercise ->
+            val interpretation = resolvedInterpretation(exercise.exerciseId) ?: return@filter false
+            (filters.movementPatternId == null || filters.movementPatternId in interpretation.patternIds) &&
+                (filters.muscleId == null || filters.muscleId in interpretation.muscles(filters.muscleRole!!)) &&
+                (filters.equipmentId == null || filters.equipmentId in exercise.equipmentIds) &&
+                (zones == null || interpretation.primaryZoneId in zones ||
+                    interpretation.secondaryZoneIds.any { it in zones })
+        }
+    }
+
+    private fun resolvedInterpretation(exerciseId: String): ExerciseInterpretation? =
+        exercisesById[exerciseId]?.takeIf {
+            it.resolutionStatus == ExerciseKnowledgeStatus.RESOLVED_FAMILY_VARIANT_LIMITED
+        }?.interpretation
+
+    companion object {
+        private val confidenceValues = KnowledgeConfidence.entries.associateBy { it.name.lowercase(Locale.ROOT) }
+        private val idPattern = Regex("[a-z][a-z0-9_]*")
+        private val exerciseIdPattern = Regex("ex_[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}")
+        private val equipmentIdPattern = Regex("(?:[a-z][a-z0-9_]*|eq_[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12})")
+        private val scientificReferenceTypes = setOf("established_anatomy", "emg_evidence", "intervention_evidence")
+
+        fun load(context: Context): TrainingKnowledgeCatalog = load(
+            context = context,
+            bodyZones = BodyZoneCatalog.load(context),
+        )
+
+        internal fun load(context: Context, bodyZones: BodyZoneCatalog): TrainingKnowledgeCatalog {
+            return load({ name -> context.assets.open(name).bufferedReader().use { it.readText() } }, bodyZones)
+        }
+
+        internal fun load(assetText: (String) -> String, bodyZones: BodyZoneCatalog): TrainingKnowledgeCatalog {
+            fun root(name: String, format: String, collection: String): JSONArray {
+                val text = assetText(name)
+                DuplicateJsonKeyValidator.validate(text)
+                val value = JSONObject(text)
+                requireKeys(value, setOf("format", "version", collection), collection)
+                check(value.getString("format") == format) { "$collection: format non pris en charge" }
+                check(value.rawInt("version") == 1) { "$collection: version non prise en charge" }
+                return value.getJSONArray(collection)
+            }
+            val referenceJson = root("science-references-v1.json", "trainlog-science-references-v1", "references")
+            val muscleJson = root("muscles-v1.json", "trainlog-muscles-v1", "muscles")
+            val actionJson = root("joint-actions-v1.json", "trainlog-joint-actions-v1", "joint_actions")
+            val patternJson = root("movement-patterns-v1.json", "trainlog-movement-patterns-v1", "movement_patterns")
+            val exerciseJson = root("exercise-knowledge-v1.json", "trainlog-exercise-knowledge-v1", "exercises")
+            val equipmentJson = root("equipment-knowledge-v1.json", "trainlog-equipment-knowledge-v1", "equipment")
+
+            val references = referenceJson.objects("ref_id") { item ->
+                requireKeys(item, setOf("ref_id", "title", "authors_or_organization", "year", "type", "url", "topics", "notes", "limitations", "doi", "pmid", "accessed_on"), "reference", setOf("publication_note"))
+                ScienceReference(item.text("ref_id"), item.text("title"), item.text("authors_or_organization"), item.nullableInt("year"), item.text("type"), item.text("url"), item.strings("topics"), item.text("notes"), item.text("limitations"), item.nullableString("doi"), item.nullableString("pmid"), item.text("accessed_on"), item.optionalString("publication_note"))
+            }
+            val refIds = references.map { it.refId }.toSet()
+            val referenceTypes = references.associate { it.refId to it.type }
+
+            data class RawMuscle(val item: JSONObject)
+            val rawMuscles = muscleJson.objects("muscle_id") { RawMuscle(it) }
+            val muscleIds = rawMuscles.map { it.item.text("muscle_id") }.toSet()
+            val actionIds = actionJson.ids("action_id")
+            val patternIds = patternJson.ids("pattern_id")
+            val zoneIds = bodyZones.zones.map { it.zoneId }.toSet()
+            val equipmentIds = equipmentJson.ids("equipment_id")
+            val exerciseIds = exerciseJson.ids("exercise_id")
+            // CONTRACT: V1 catalogs are additive; identity, ordering, and cross-reference
+            // validation below define integrity without freezing today's collection sizes.
+
+            val muscles = rawMuscles.map { raw ->
+                val item = raw.item
+                val required = setOf("muscle_id", "display_name", "display_name_fr", "entity_type", "anatomical_group", "aggregate_group_id", "body_zone_id", "body_zone_ids", "joint_action_ids", "primary_actions", "primary_actions_semantics", "overlap_warning", "functional_notes", "confidence", "evidence_type", "source_refs")
+                requireKeys(item, required, "muscle", setOf("member_muscle_ids"))
+                val aggregate = item.nullableString("aggregate_group_id")
+                check(aggregate == null || aggregate in muscleIds) { "aggregate_group_id pendant" }
+                val members = item.optionalStrings("member_muscle_ids")
+                check(members.all { it in muscleIds }) { "membre musculaire pendant" }
+                val zones = item.strings("body_zone_ids"); val actions = item.strings("joint_action_ids"); val primaryActions = item.strings("primary_actions"); val refs = item.strings("source_refs", true)
+                check(item.text("body_zone_id") in zoneIds && zones.all { it in zoneIds } && actions.all { it in actionIds } && primaryActions.all { it in actionIds } && refs.all { it in refIds }) { "référence croisée muscle invalide" }
+                MuscleKnowledge(item.text("muscle_id"), item.text("display_name"), item.text("display_name_fr"), enumValue(item.text("entity_type"), MuscleEntityType.entries), item.text("anatomical_group"), aggregate, item.text("body_zone_id"), zones, actions, primaryActions, item.text("primary_actions_semantics"), item.text("overlap_warning"), item.string("functional_notes"), confidence(item), item.text("evidence_type"), refs, members)
+            }
+
+            val actions = actionJson.objects("action_id") { item ->
+                requireKeys(item, setOf("action_id", "display_name_fr", "definition", "anatomical_region", "joint_complex", "joint_or_complex", "principal_plane", "plane_notes", "contributing_muscle_ids", "contributor_semantics", "evidence_type", "confidence", "source_refs", "notes"), "joint_action")
+                val contributors = item.strings("contributing_muscle_ids"); val refs = item.strings("source_refs", true)
+                check(contributors.all { it in muscleIds } && refs.all { it in refIds }) { "référence croisée action invalide" }
+                JointActionKnowledge(item.text("action_id"), item.text("display_name_fr"), item.text("definition"), item.text("anatomical_region"), item.text("joint_complex"), item.text("joint_or_complex"), item.text("principal_plane"), item.text("plane_notes"), contributors, item.text("contributor_semantics"), item.text("evidence_type"), confidence(item), refs, item.text("notes"))
+            }
+            val patterns = patternJson.objects("pattern_id") { item ->
+                val required = setOf("pattern_id", "display_name_fr", "definition", "typical_action_ids", "typical_body_zone_ids", "evidence_type", "confidence", "source_refs", "notes")
+                requireKeys(item, required, "movement_pattern", setOf("body_zone_semantics", "parent_pattern_id"))
+                val typicalActions = item.strings("typical_action_ids"); val typicalZones = item.strings("typical_body_zone_ids"); val refs = item.strings("source_refs", true)
+                check(typicalActions.all { it in actionIds } && typicalZones.all { it in zoneIds } && refs.all { it in refIds }) { "référence croisée pattern invalide" }
+                val parent = item.optionalString("parent_pattern_id"); check(parent == null || parent in patternIds) { "parent_pattern_id pendant" }
+                MovementPatternKnowledge(item.text("pattern_id"), item.text("display_name_fr"), item.text("definition"), typicalActions, typicalZones, item.text("evidence_type"), confidence(item), refs, item.text("notes"), item.optionalString("body_zone_semantics"), parent)
+            }
+            fun interpretation(item: JSONObject, conditional: Boolean): ExerciseInterpretation {
+                val required = setOf("family_description", "action_ids", "pattern_ids", "primary_muscle_ids", "secondary_muscle_ids", "stabilizer_muscle_ids", "primary_zone_id", "secondary_zone_ids", "confidence", "evidence_type", "source_refs", "variant_notes", "role_notes")
+                requireKeys(item, required, "interpretation", if (conditional) setOf("required_confirmation") else emptySet())
+                val actionRefs = item.strings("action_ids"); val patternRefs = item.strings("pattern_ids")
+                val primary = item.strings("primary_muscle_ids"); val secondary = item.strings("secondary_muscle_ids"); val stabilizers = item.strings("stabilizer_muscle_ids")
+                check((primary + secondary + stabilizers).size == (primary + secondary + stabilizers).toSet().size) { "muscle présent dans plusieurs rôles" }
+                val secondaryZones = item.strings("secondary_zone_ids"); val refs = item.strings("source_refs", true); val primaryZone = item.text("primary_zone_id")
+                check(actionRefs.all { it in actionIds } && patternRefs.all { it in patternIds } && (primary + secondary + stabilizers).all { it in muscleIds } && primaryZone in zoneIds && secondaryZones.all { it in zoneIds } && primaryZone !in secondaryZones && refs.all { it in refIds }) { "référence croisée interprétation invalide" }
+                return ExerciseInterpretation(item.text("family_description"), actionRefs, patternRefs, primary, secondary, stabilizers, primaryZone, secondaryZones, confidence(item), item.text("evidence_type"), refs, item.text("variant_notes"), item.text("role_notes"), item.optionalString("required_confirmation"))
+            }
+            val exercises = exerciseJson.objects("exercise_id") { item ->
+                val required = setOf("exercise_id", "exercise_name", "equipment_ids", "identity_evidence", "identity_evidence_type", "resolution_status", "confidence", "interpretation", "conditional_interpretation", "existing_body_zones", "source_refs", "limitations", "body_zone_audit")
+                requireKeys(item, required, "exercise", setOf("equipment_link_status"))
+                val status = enumValue(item.text("resolution_status"), ExerciseKnowledgeStatus.entries)
+                val regular = item.nullableObject("interpretation")?.let { interpretation(it, false) }
+                val conditional = item.nullableObject("conditional_interpretation")?.let { interpretation(it, true) }
+                check((status == ExerciseKnowledgeStatus.RESOLVED_FAMILY_VARIANT_LIMITED && regular != null && conditional == null) || (status == ExerciseKnowledgeStatus.CONDITIONAL && regular == null && conditional != null) || (status == ExerciseKnowledgeStatus.UNRESOLVED && regular == null && conditional == null)) { "interprétation incompatible avec resolution_status" }
+                val equipmentRefs = item.strings("equipment_ids"); val refs = item.strings("source_refs")
+                check(status == ExerciseKnowledgeStatus.UNRESOLVED || refs.isNotEmpty()) { "exercice résolu/conditionnel sans source" }
+                check(equipmentRefs.all { it in equipmentIds } && refs.all { it in refIds }) { "référence croisée exercice invalide" }
+                val existing = item.getJSONObject("existing_body_zones"); requireKeys(existing, setOf("primary_zone_id", "secondary_zone_ids"), "existing_body_zones")
+                val existingZones = ExistingBodyZoneMapping(existing.nullableString("primary_zone_id"), existing.strings("secondary_zone_ids")); check((existingZones.primaryZoneId == null || existingZones.primaryZoneId in zoneIds) && existingZones.secondaryZoneIds.all { it in zoneIds }) { "zone existante pendante" }
+                val audit = item.getJSONObject("body_zone_audit")
+                requireKeys(audit, setOf("status", "severity", "confidence", "rationale", "existing_primary_zone_id", "existing_secondary_zone_ids", "proposed_mutation", "source_refs"), "body_zone_audit")
+                val auditPrimary = audit.nullableString("existing_primary_zone_id")
+                val auditSecondary = audit.strings("existing_secondary_zone_ids")
+                val auditRefs = audit.strings("source_refs")
+                check((auditPrimary == null || auditPrimary in zoneIds) && auditSecondary.all { it in zoneIds }) { "zone d'audit pendante" }
+                check(auditRefs.all { it in refIds }) { "référence d'audit pendante" }
+                // INVARIANT: every non-unresolved BODY ZONE conclusion remains traceable
+                // to evidence, matching the canonical cross-catalog validator.
+                check(audit.text("status") == "unresolved" || auditRefs.isNotEmpty()) { "audit BODY ZONE résolu sans source" }
+                val bodyZoneAudit = BodyZoneAudit(
+                    enumValue(audit.text("status"), BodyZoneAuditStatus.entries),
+                    enumValue(audit.text("severity"), BodyZoneAuditSeverity.entries),
+                    confidence(audit), audit.text("rationale"), auditPrimary, auditSecondary,
+                    audit.nullableString("proposed_mutation"), auditRefs,
+                )
+                ExerciseKnowledge(item.text("exercise_id"), item.text("exercise_name"), equipmentRefs, item.text("identity_evidence"), item.text("identity_evidence_type"), status, confidence(item), regular, conditional, existingZones, refs, item.strings("limitations"), item.optionalString("equipment_link_status"), bodyZoneAudit)
+            }
+            val linked = mutableSetOf<Pair<String, String>>()
+            val equipment = equipmentJson.objects("equipment_id") { item ->
+                val required = setOf("equipment_id", "manufacturer", "model", "identification_status", "scientific_status", "catalog_type", "catalog_load_semantics", "mechanics", "evidence_type", "confidence", "source_refs", "capabilities", "requires_actual_exercise", "limitations", "scientific_status_scope")
+                requireKeys(item, required, "equipment", setOf("audit_note"))
+                val refs = item.strings("source_refs", true); check(refs.all { it in refIds }) { "référence équipement pendante" }
+                val conf = confidence(item); check(!(conf == KnowledgeConfidence.HIGH && item.text("evidence_type") == "manufacturer_statement")) { "cartographie anatomique élevée fondée seulement sur fabricant" }
+                // CONTRACT: HIGH equipment mappings require at least one admissible
+                // scientific source type; Android must accept exactly the validator policy.
+                check(conf != KnowledgeConfidence.HIGH || refs.any { referenceTypes[it] in scientificReferenceTypes }) { "cartographie équipement élevée sans source scientifique" }
+                val capabilities = item.getJSONArray("capabilities").objectsUnordered { capability ->
+                    requireKeys(capability, setOf("display_name", "exercise_ids", "link_status", "interpretation", "requirements"), "capability")
+                    val capabilityExercises = capability.strings("exercise_ids"); check(capabilityExercises.all { it in exerciseIds }) { "exercice fantôme dans capability" }
+                    check(capability.text("link_status") in setOf("linked_existing", "linked_identity_conditional_interpretation", "unlinked_capability", "catalog_compatible_not_observed_occurrence")) { "link_status invalide" }
+                    capabilityExercises.forEach { linked += it to item.text("equipment_id") }
+                    // WHY: a one-sided capability would make desktop and Android
+                    // compatibility queries disagree over the same immutable assets.
+                    capabilityExercises.forEach { exerciseId ->
+                        check(item.text("equipment_id") in exercises.single { it.exerciseId == exerciseId }.equipmentIds) { "compatibilité équipement/exercice asymétrique" }
+                    }
+                    EquipmentCapabilityKnowledge(capability.text("display_name"), capabilityExercises, capability.text("link_status"), capability.nullableObject("interpretation")?.let { interpretation(it, false) }, capability.text("requirements"))
+                }
+                EquipmentKnowledge(item.text("equipment_id"), item.nullableString("manufacturer"), item.nullableString("model"), item.text("identification_status"), enumValue(item.text("scientific_status"), EquipmentScienceStatus.entries), item.text("catalog_type"), item.nullableString("catalog_load_semantics")?.let { enumValue(it, EquipmentLoadSemantics.entries) }, item.text("mechanics"), item.text("evidence_type"), conf, refs, capabilities, item.rawBoolean("requires_actual_exercise"), item.strings("limitations"), item.optionalString("audit_note"), item.text("scientific_status_scope"))
+            }
+            exercises.forEach { exercise -> exercise.equipmentIds.forEach { check(exercise.exerciseId to it in linked) { "compatibilité exercice/équipement asymétrique" } } }
+            return TrainingKnowledgeCatalog(references, muscles, actions, patterns, exercises, equipment, bodyZones)
+        }
+
+        private fun confidence(item: JSONObject): KnowledgeConfidence = confidenceValues[item.text("confidence")] ?: error("confidence invalide")
+        private fun <T : Enum<T>> enumValue(value: String, entries: List<T>): T = entries.firstOrNull { it.name.lowercase(Locale.ROOT) == value } ?: error("valeur enum invalide: $value")
+        private fun requireKeys(value: JSONObject, required: Set<String>, where: String, optional: Set<String> = emptySet()) {
+            val keys = value.keys().asSequence().toSet(); check(required.all { it in keys } && keys.all { it in required || it in optional }) { "$where: clés invalides" }
+        }
+        private fun JSONObject.text(key: String): String = get(key).let { check(it is String && it.isNotBlank()) { "$key: texte requis" }; it }
+        private fun JSONObject.string(key: String): String = get(key).let { check(it is String) { "$key: chaîne attendue" }; it }
+        private fun JSONObject.nullableString(key: String): String? = if (isNull(key)) null else get(key).let { check(it is String) { "$key: chaîne attendue" }; it }
+        private fun JSONObject.optionalString(key: String): String? = if (!has(key) || isNull(key)) null else nullableString(key)
+        private fun JSONObject.rawInt(key: String): Int = get(key).let { check(it is Int) { "$key: entier attendu" }; it }
+        private fun JSONObject.nullableInt(key: String): Int? = if (isNull(key)) null else rawInt(key)
+        private fun JSONObject.rawBoolean(key: String): Boolean = get(key).let { check(it is Boolean) { "$key: booléen attendu" }; it }
+        private fun JSONObject.nullableObject(key: String): JSONObject? = if (isNull(key)) null else get(key).let { check(it is JSONObject) { "$key: objet attendu" }; it }
+        private fun JSONObject.strings(key: String, nonempty: Boolean = false): List<String> = get(key).let { value ->
+            check(value is JSONArray) { "$key: tableau attendu" }; List(value.length()) { index -> value.get(index).let { check(it is String && it.isNotBlank()) { "$key: chaîne vide/invalide" }; it } }.also {
+                check(it.size == it.toSet().size && (!nonempty || it.isNotEmpty())) { "$key: doublon ou tableau vide" }
+                if (key.endsWith("_ids") || key == "source_refs" || key == "joint_action_ids" || key == "primary_actions") {
+                    check(it == it.sorted()) { "$key: ordre stable requis" }
+                }
+            }
+        }
+        private fun JSONObject.optionalStrings(key: String): List<String> = if (has(key)) strings(key) else emptyList()
+        private fun JSONArray.ids(key: String): Set<String> = objects(key) { it.text(key) }.toSet()
+        private fun <T> JSONArray.objects(key: String, transform: (JSONObject) -> T): List<T> {
+            val previous = mutableSetOf<String>(); var last: String? = null
+            return objectsUnordered { item ->
+                val id = item.text(key)
+                val validId = when (key) {
+                    "exercise_id" -> exerciseIdPattern.matches(id)
+                    "equipment_id" -> equipmentIdPattern.matches(id)
+                    else -> idPattern.matches(id)
+                }
+                check(validId) { "$key invalide" }
+                check(previous.add(id) && (last == null || last!! < id)) { "$key dupliqué ou hors ordre" }
+                last = id
+                transform(item)
+            }
+        }
+        private fun <T> JSONArray.objectsUnordered(transform: (JSONObject) -> T): List<T> = List(length()) { index -> get(index).let { check(it is JSONObject) { "objet attendu" }; transform(it) } }
+    }
+}
+
+/** Small lexical pass because org.json otherwise silently accepts duplicate object keys. */
+private object DuplicateJsonKeyValidator {
+    fun validate(text: String) { Parser(text).parse() }
+    private class Parser(private val source: String) {
+        private var index = 0
+        fun parse() { value(); whitespace(); check(index == source.length) { "JSON suffixe invalide" } }
+        private fun value() { whitespace(); check(index < source.length) { "JSON tronqué" }; when (source[index]) { '{' -> objectValue(); '[' -> arrayValue(); '"' -> string(); 't' -> literal("true"); 'f' -> literal("false"); 'n' -> literal("null"); else -> number() } }
+        private fun objectValue() { index++; whitespace(); val keys = mutableSetOf<String>(); if (take('}')) return; while (true) { whitespace(); val key = string(); check(keys.add(key)) { "clé JSON dupliquée: $key" }; whitespace(); expect(':'); value(); whitespace(); if (take('}')) return; expect(',') } }
+        private fun arrayValue() { index++; whitespace(); if (take(']')) return; while (true) { value(); whitespace(); if (take(']')) return; expect(',') } }
+        private fun string(): String { expect('"'); val out = StringBuilder(); while (index < source.length) { val c = source[index++]; when (c) { '"' -> return out.toString(); '\\' -> { check(index < source.length); val escaped = source[index++]; if (escaped == 'u') { check(index + 4 <= source.length); val hex = source.substring(index, index + 4); check(hex.all { it.isDigit() || it.lowercaseChar() in 'a'..'f' }); out.append(hex.toInt(16).toChar()); index += 4 } else { check(escaped in "\"\\/bfnrt"); out.append(escaped) } }; else -> { check(c.code >= 0x20); out.append(c) } } }; error("chaîne JSON tronquée") }
+        private fun literal(value: String) { check(source.startsWith(value, index)); index += value.length }
+        private fun number() { val start = index; if (take('-')) Unit; check(index < source.length && source[index].isDigit()); if (source[index] == '0') index++ else while (index < source.length && source[index].isDigit()) index++; if (take('.')) { check(index < source.length && source[index].isDigit()); while (index < source.length && source[index].isDigit()) index++ }; if (index < source.length && source[index] in "eE") { index++; if (index < source.length && source[index] in "+-") index++; check(index < source.length && source[index].isDigit()); while (index < source.length && source[index].isDigit()) index++ }; check(index > start) }
+        private fun whitespace() { while (index < source.length && source[index].isWhitespace()) index++ }
+        private fun take(c: Char): Boolean { if (index < source.length && source[index] == c) { index++; return true }; return false }
+        private fun expect(c: Char) { check(take(c)) { "JSON: '$c' attendu à $index" } }
+    }
+}

android/app/src/main/java/com/labfytools/trainlog/data/TrainlogRepository.kt

@@ -184,6 +184,68 @@ sealed interface FinalizeActiveDraftResult {
     ) : FinalizeActiveDraftResult
 }
 
+data class ExerciseOccurrenceCursor(
+    val startedAt: String,
+    val sessionId: String,
+    val entryId: String,
+)
+
+data class ExerciseSetContext(
+    val position: Int,
+    val reps: Int?,
+    val durationSeconds: Int?,
+    /** Null and 0.0 are intentionally distinct actual observations. */
+    val weightKg: Double?,
+)
+
+data class ExerciseSetPage(
+    val sets: List<ExerciseSetContext>,
+    val nextPosition: Int?,
+)
+
+data class ExerciseOccurrenceContext(
+    val sessionId: String,
+    val entryId: String,
+    val startedAt: String,
+    val sessionType: SessionType,
+    val equipmentId: String?,
+    val equipmentDisplayName: String?,
+    val loadSemantics: EquipmentLoadSemantics?,
+    val recordingMode: RecordingMode,
+    val trackingMode: TrackingMode,
+    val dataFields: Int,
+    val setPreview: ExerciseSetPage?,
+    val continuousDurationSeconds: Int?,
+    val speedKmh: Double?,
+    val distanceKm: Double?,
+)
+
+data class ExerciseOccurrencePage(
+    val occurrences: List<ExerciseOccurrenceContext>,
+    val nextCursor: ExerciseOccurrenceCursor?,
+)
+
+data class ExplicitMaxContext(
+    val sessionId: String,
+    val entryId: String,
+    val startedAt: String,
+    val maxWeightKg: Double,
+    val equipmentId: String?,
+    val equipmentDisplayName: String?,
+    val loadSemantics: EquipmentLoadSemantics?,
+)
+
+data class TrainingExerciseContext(
+    val exercise: ExerciseProfile,
+    /** Persisted user classification; never replaced by scientific projection. */
+    val persistedDirectZoneIds: List<String>,
+    val persistedZoneIdsWithAncestors: List<String>,
+    val knowledge: ExerciseKnowledge?,
+    val compatibleEquipment: List<EquipmentKnowledge>,
+    val latestExplicitMax: ExplicitMaxContext?,
+    val recentPerformance: ExerciseOccurrencePage,
+)
+
 class TrainlogRepository(
     context: Context,
     databaseName: String =
@@ -191,6 +253,7 @@ class TrainlogRepository(
 ) {
     private val applicationContext = context.applicationContext
     private val bodyZones = BodyZoneCatalog.load(applicationContext)
+    private val trainingKnowledge = TrainingKnowledgeCatalog.load(applicationContext, bodyZones)
     private val database =
         TrainlogDatabaseHelper(
             applicationContext,
@@ -201,6 +264,23 @@ class TrainlogRepository(
         database.close()
     }
 
+    fun getExerciseKnowledge(exerciseId: String): ExerciseKnowledge? =
+        trainingKnowledge.getExerciseKnowledge(exerciseId)
+
+    fun getConditionalExerciseKnowledge(exerciseId: String): ExerciseKnowledge? =
+        trainingKnowledge.getConditionalExerciseKnowledge(exerciseId)
+
+    fun getMuscleKnowledge(muscleId: String): MuscleKnowledge? = trainingKnowledge.getMuscle(muscleId)
+    fun getJointActionKnowledge(actionId: String): JointActionKnowledge? = trainingKnowledge.getJointAction(actionId)
+    fun getMovementPatternKnowledge(patternId: String): MovementPatternKnowledge? = trainingKnowledge.getMovementPattern(patternId)
+    fun getScienceReference(refId: String): ScienceReference? = trainingKnowledge.getReference(refId)
+    fun getEquipmentKnowledge(equipmentId: String): EquipmentKnowledge? = trainingKnowledge.getEquipmentKnowledge(equipmentId)
+    fun getScientificBodyZoneMapping(exerciseId: String): ScientificBodyZoneMapping? = trainingKnowledge.getScientificBodyZoneMapping(exerciseId)
+    fun listExercisesByMovementPattern(patternId: String): List<ExerciseKnowledge> = trainingKnowledge.listExercisesByMovementPattern(patternId)
+    fun listExercisesByMuscle(muscleId: String, role: MuscleRole): List<ExerciseKnowledge> = trainingKnowledge.listExercisesByMuscle(muscleId, role)
+    fun listCompatibleKnowledgeExercises(equipmentId: String): List<ExerciseKnowledge> = trainingKnowledge.listCompatibleExercises(equipmentId)
+    fun queryExerciseKnowledge(filters: KnowledgeExerciseFilters): List<ExerciseKnowledge> = trainingKnowledge.queryExercises(filters)
+
     fun listEquipment(): List<EquipmentCatalogEntry> {
         val output = mutableListOf<EquipmentCatalogEntry>()
         database.readableDatabase.query(
@@ -3172,6 +3252,286 @@ class TrainlogRepository(
         )
     }
 
+    /**
+     * Compose immutable scientific metadata with exact persisted runtime state.
+     * WHY: names are editable labels, so every join and lookup stays on stable
+     * exercise_id; the transaction gives all components one SQLite read view.
+     */
+    fun getTrainingExerciseContext(
+        exerciseId: String,
+        occurrenceLimit: Int = 8,
+        setPreviewLimit: Int = 8,
+    ): TrainingExerciseContext? {
+        require(occurrenceLimit in 1..MAX_OCCURRENCE_PAGE_SIZE) { "occurrenceLimit doit être compris entre 1 et $MAX_OCCURRENCE_PAGE_SIZE" }
+        require(setPreviewLimit in 1..MAX_SET_PAGE_SIZE) { "setPreviewLimit doit être compris entre 1 et $MAX_SET_PAGE_SIZE" }
+        val db = database.readableDatabase
+        val ownsTransaction = !db.inTransaction()
+        if (ownsTransaction) db.beginTransactionNonExclusive()
+        return try {
+            val exercise = readExerciseProfileExact(db, exerciseId) ?: return null
+            val direct = buildList {
+                exercise.primaryZoneId?.let(::add)
+                addAll(exercise.secondaryZoneIds)
+            }
+            val expanded = linkedSetOf<String>()
+            direct.forEach { zoneId ->
+                expanded += zoneId
+                expanded += bodyZones.ancestors(zoneId).map { it.zoneId }
+            }
+            val knowledge = trainingKnowledge.getExerciseKnowledge(exerciseId)
+            val compatible = knowledge?.equipmentIds.orEmpty().mapNotNull(trainingKnowledge::getEquipmentKnowledge)
+            val result = TrainingExerciseContext(
+                exercise = exercise,
+                persistedDirectZoneIds = direct,
+                persistedZoneIdsWithAncestors = bodyZones.zones.map { it.zoneId }.filter { it in expanded },
+                knowledge = knowledge,
+                compatibleEquipment = compatible,
+                latestExplicitMax = readLatestExplicitMax(db, exerciseId),
+                recentPerformance = readExerciseOccurrencePage(db, exerciseId, occurrenceLimit, null, setPreviewLimit),
+            )
+            if (ownsTransaction) db.setTransactionSuccessful()
+            result
+        } finally {
+            if (ownsTransaction) db.endTransaction()
+        }
+    }
+
+    /** Deterministic keyset page over current data; pages do not hold a cross-call snapshot. */
+    fun listExerciseOccurrences(
+        exerciseId: String,
+        limit: Int,
+        after: ExerciseOccurrenceCursor? = null,
+        setPreviewLimit: Int = 8,
+    ): ExerciseOccurrencePage {
+        require(limit in 1..MAX_OCCURRENCE_PAGE_SIZE) { "limit doit être compris entre 1 et $MAX_OCCURRENCE_PAGE_SIZE" }
+        require(setPreviewLimit in 1..MAX_SET_PAGE_SIZE) { "setPreviewLimit doit être compris entre 1 et $MAX_SET_PAGE_SIZE" }
+        validateOccurrenceCursor(after)
+        val db = database.readableDatabase
+        val ownsTransaction = !db.inTransaction()
+        if (ownsTransaction) db.beginTransactionNonExclusive()
+        return try {
+            require(readExerciseProfileExact(db, exerciseId) != null) { "exercise_id inconnu: $exerciseId" }
+            val result = readExerciseOccurrencePage(db, exerciseId, limit, after, setPreviewLimit)
+            if (ownsTransaction) db.setTransactionSuccessful()
+            result
+        } finally {
+            if (ownsTransaction) db.endTransaction()
+        }
+    }
+
+    /** Follow-up bounded set page for one exact occurrence; no global set load exists. */
+    fun listExerciseOccurrenceSets(
+        exerciseId: String,
+        entryId: String,
+        limit: Int,
+        afterPosition: Int? = null,
+    ): ExerciseSetPage {
+        require(exerciseId.isNotBlank() && entryId.isNotBlank()) { "identités vides" }
+        require(limit in 1..MAX_SET_PAGE_SIZE) { "limit doit être compris entre 1 et $MAX_SET_PAGE_SIZE" }
+        require(afterPosition == null || afterPosition >= 0) { "position de curseur invalide" }
+        val db = database.readableDatabase
+        val ownsTransaction = !db.inTransaction()
+        if (ownsTransaction) db.beginTransactionNonExclusive()
+        return try {
+            val rowId = db.rawQuery(
+                "SELECT se.id,se.recording_mode FROM session_exercises se JOIN exercises e ON e.id=se.exercise_row_id WHERE e.exercise_id=? AND se.entry_id=?;",
+                arrayOf(exerciseId, entryId),
+            ).use { cursor ->
+                require(cursor.moveToFirst()) { "occurrence inconnue pour cet exercice" }
+                require(cursor.getString(1) == "sets") { "une activité continue ne possède pas de séries" }
+                cursor.getLong(0)
+            }
+            val result = readSetPage(db, rowId, limit, afterPosition)
+            if (ownsTransaction) db.setTransactionSuccessful()
+            result
+        } finally {
+            if (ownsTransaction) db.endTransaction()
+        }
+    }
+
+    private fun readExerciseProfileExact(db: SQLiteDatabase, exerciseId: String): ExerciseProfile? =
+        db.rawQuery(
+            "SELECT exercise_id,name,normalized_name,recording_mode,tracking_mode,data_fields FROM exercises WHERE exercise_id=?;",
+            arrayOf(exerciseId),
+        ).use { cursor ->
+            if (!cursor.moveToFirst()) null else {
+                val zones = readExerciseBodyZones(db, exerciseId)
+                ExerciseProfile(
+                    exerciseId = cursor.getString(0), name = cursor.getString(1), normalizedName = cursor.getString(2),
+                    recordingMode = parseRecordingMode(cursor.getString(3)), trackingMode = parseTrackingMode(cursor.getString(4)),
+                    dataFields = checkedNonNegativeInt(cursor.getLong(5), "data_fields"), primaryZoneId = zones.first,
+                    secondaryZoneIds = zones.second,
+                )
+            }
+        }
+
+    private data class TemporalCandidate(
+        val rowId: Long,
+        val sessionId: String,
+        val entryId: String,
+        val startedAt: String,
+        val timestamp: TrainlogTimestampKey,
+    )
+
+    private fun compareTemporal(left: TemporalCandidate, right: TemporalCandidate): Int =
+        left.timestamp.compareTo(right.timestamp).takeIf { it != 0 }
+            ?: TrainlogTimestamp.compareIds(left.sessionId, right.sessionId).takeIf { it != 0 }
+            ?: TrainlogTimestamp.compareIds(left.entryId, right.entryId)
+
+    private fun candidate(cursor: android.database.Cursor): TemporalCandidate {
+        val startedAt = cursor.requiredText(3, "started_at")
+        return TemporalCandidate(
+            rowId = cursor.getLong(0),
+            sessionId = cursor.requiredText(1, "session_id"),
+            entryId = cursor.requiredText(2, "entry_id"),
+            startedAt = startedAt,
+            timestamp = checkNotNull(TrainlogTimestamp.parse(startedAt)) {
+                "started_at persistant invalide pour l'occurrence"
+            },
+        )
+    }
+
+    private fun retainCandidate(rows: MutableList<TemporalCandidate>, value: TemporalCandidate, capacity: Int) {
+        val position = rows.indexOfFirst { compareTemporal(value, it) > 0 }.let { if (it < 0) rows.size else it }
+        if (position < capacity) {
+            rows.add(position, value)
+            if (rows.size > capacity) rows.removeAt(rows.lastIndex)
+        }
+    }
+
+    private fun readLatestExplicitMax(db: SQLiteDatabase, exerciseId: String): ExplicitMaxContext? {
+        val selected = mutableListOf<TemporalCandidate>()
+        db.rawQuery(
+            """SELECT se.id,s.session_id,se.entry_id,s.started_at FROM max_results mr
+               JOIN session_exercises se ON se.id=mr.session_exercise_row_id
+               JOIN sessions s ON s.id=se.session_row_id JOIN exercises e ON e.id=se.exercise_row_id
+               WHERE e.exercise_id=? AND s.session_type='max_test';""",
+            arrayOf(exerciseId),
+        ).use { cursor -> while (cursor.moveToNext()) retainCandidate(selected, candidate(cursor), 1) }
+        val winner = selected.singleOrNull() ?: return null
+        return db.rawQuery(
+            """SELECT s.session_id,se.entry_id,s.started_at,mr.max_weight_kg,
+                      eq.equipment_id,eq.display_name,eq.load_semantics
+               FROM max_results mr JOIN session_exercises se ON se.id=mr.session_exercise_row_id
+               JOIN sessions s ON s.id=se.session_row_id LEFT JOIN equipment eq ON eq.id=se.equipment_row_id
+               WHERE se.id=?;""",
+            arrayOf(winner.rowId.toString()),
+        ).use { cursor ->
+            check(cursor.moveToFirst()) { "résultat MAX sélectionné absent" }
+            ExplicitMaxContext(
+                sessionId = cursor.requiredText(0, "session_id"), entryId = cursor.requiredText(1, "entry_id"),
+                startedAt = cursor.requiredText(2, "started_at"), maxWeightKg = cursor.finiteNonNegativeDouble(3, "max_weight_kg", strictlyPositive = true),
+                equipmentId = cursor.optionalText(4), equipmentDisplayName = cursor.optionalText(5),
+                loadSemantics = cursor.optionalText(6)?.let(::parseLoadSemantics),
+            )
+        }
+    }
+
+    private fun readExerciseOccurrencePage(
+        db: SQLiteDatabase, exerciseId: String, limit: Int, after: ExerciseOccurrenceCursor?, setPreviewLimit: Int,
+    ): ExerciseOccurrencePage {
+        val afterCandidate = after?.let {
+            TemporalCandidate(-1L, it.sessionId, it.entryId, it.startedAt,
+                requireNotNull(TrainlogTimestamp.parse(it.startedAt)))
+        }
+        val selected = mutableListOf<TemporalCandidate>()
+        db.rawQuery(
+            """
+            SELECT se.id,s.session_id,se.entry_id,s.started_at
+            FROM session_exercises se
+            JOIN sessions s ON s.id=se.session_row_id
+            JOIN exercises e ON e.id=se.exercise_row_id
+            WHERE e.exercise_id=?;
+            """.trimIndent(), arrayOf(exerciseId),
+        ).use { cursor ->
+            while (cursor.moveToNext()) {
+                val value = candidate(cursor)
+                // INVARIANT: selection and the exclusive cursor share compareTemporal.
+                if (afterCandidate == null || compareTemporal(value, afterCandidate) < 0)
+                    retainCandidate(selected, value, limit + 1)
+            }
+        }
+        val hasMore = selected.size > limit
+        val kept = if (hasMore) selected.take(limit) else selected
+        val rows = kept.map { selectedRow ->
+            db.rawQuery(
+                """SELECT se.id,s.session_id,se.entry_id,s.started_at,s.session_type,
+                          eq.equipment_id,eq.display_name,eq.load_semantics,
+                          se.recording_mode,se.tracking_mode,se.data_fields,
+                          ca.duration_seconds,ca.speed_kmh,ca.distance_km
+                   FROM session_exercises se JOIN sessions s ON s.id=se.session_row_id
+                   LEFT JOIN equipment eq ON eq.id=se.equipment_row_id
+                   LEFT JOIN continuous_activity ca ON ca.session_exercise_row_id=se.id
+                   WHERE se.id=?;""",
+                arrayOf(selectedRow.rowId.toString()),
+            ).use { cursor ->
+                check(cursor.moveToFirst()) { "occurrence sélectionnée absente" }
+                val recording = parseRecordingMode(cursor.requiredText(8, "recording_mode"))
+                val rowId = cursor.getLong(0)
+                val context = ExerciseOccurrenceContext(
+                    sessionId = cursor.requiredText(1, "session_id"), entryId = cursor.requiredText(2, "entry_id"),
+                    startedAt = cursor.requiredText(3, "started_at"), sessionType = SessionType.fromWire(cursor.requiredText(4, "session_type")),
+                    equipmentId = cursor.optionalText(5), equipmentDisplayName = cursor.optionalText(6),
+                    loadSemantics = cursor.optionalText(7)?.let(::parseLoadSemantics), recordingMode = recording,
+                    trackingMode = parseTrackingMode(cursor.requiredText(9, "tracking_mode")), dataFields = checkedNonNegativeInt(cursor.getLong(10), "data_fields"),
+                    setPreview = if (recording == RecordingMode.SETS) readSetPage(db, rowId, setPreviewLimit, null) else null,
+                    continuousDurationSeconds = if (recording == RecordingMode.CONTINUOUS) cursor.requiredPositiveInt(11, "duration_seconds") else null,
+                    speedKmh = cursor.optionalFinitePositiveDouble(12, "speed_kmh"), distanceKm = cursor.optionalFinitePositiveDouble(13, "distance_km"),
+                )
+                if (recording == RecordingMode.SETS) check(cursor.isNull(11) && cursor.isNull(12) && cursor.isNull(13)) { "activité continue attachée à une occurrence SETS" }
+                context
+            }
+        }
+        val last = rows.lastOrNull()
+        return ExerciseOccurrencePage(
+            occurrences = rows,
+            nextCursor = if (hasMore && last != null) ExerciseOccurrenceCursor(last.startedAt, last.sessionId, last.entryId) else null,
+        )
+    }
+
+    private fun readSetPage(db: SQLiteDatabase, occurrenceRowId: Long, limit: Int, afterPosition: Int?): ExerciseSetPage {
+        val selection = if (afterPosition == null) "session_exercise_row_id=?" else "session_exercise_row_id=? AND position>?"
+        val args = if (afterPosition == null) arrayOf(occurrenceRowId.toString(), (limit + 1).toString()) else arrayOf(occurrenceRowId.toString(), afterPosition.toString(), (limit + 1).toString())
+        val rows = mutableListOf<ExerciseSetContext>()
+        db.rawQuery("SELECT position,reps,duration_seconds,weight_kg FROM performed_sets WHERE $selection ORDER BY position ASC LIMIT ?;", args).use { cursor ->
+            while (cursor.moveToNext()) {
+                val reps = if (cursor.isNull(1)) null else checkedNonNegativeInt(cursor.getLong(1), "reps")
+                val duration = if (cursor.isNull(2)) null else cursor.requiredPositiveInt(2, "duration_seconds")
+                check((reps == null) != (duration == null)) { "forme de série corrompue" }
+                rows += ExerciseSetContext(checkedNonNegativeInt(cursor.getLong(0), "position"), reps, duration, if (cursor.isNull(3)) null else cursor.finiteNonNegativeDouble(3, "weight_kg"))
+            }
+        }
+        val hasMore = rows.size > limit
+        val kept = if (hasMore) rows.take(limit) else rows
+        return ExerciseSetPage(kept, if (hasMore) kept.last().position else null)
+    }
+
+    private fun validateOccurrenceCursor(cursor: ExerciseOccurrenceCursor?) {
+        if (cursor == null) return
+        require(cursor.startedAt.isNotBlank() && cursor.sessionId.isNotBlank() && cursor.entryId.isNotBlank()) { "curseur d'occurrence invalide" }
+        require(TrainlogTimestamp.parse(cursor.startedAt) != null) { "startedAt du curseur invalide" }
+    }
+
+    private fun parseRecordingMode(value: String): RecordingMode = when (value) {
+        "sets" -> RecordingMode.SETS; "continuous" -> RecordingMode.CONTINUOUS; else -> error("recording_mode corrompu: $value")
+    }
+    private fun parseTrackingMode(value: String): TrackingMode = when (value) {
+        "reps" -> TrackingMode.REPS; "duration" -> TrackingMode.DURATION; else -> error("tracking_mode corrompu: $value")
+    }
+    private fun parseLoadSemantics(value: String): EquipmentLoadSemantics = runCatching { EquipmentLoadSemantics.valueOf(value.uppercase(Locale.ROOT)) }.getOrElse { error("load_semantics corrompu: $value") }
+
+    private fun android.database.Cursor.requiredText(index: Int, name: String): String = getString(index)?.takeIf { it.isNotBlank() } ?: error("$name absent")
+    private fun android.database.Cursor.optionalText(index: Int): String? = if (isNull(index)) null else getString(index)
+    private fun android.database.Cursor.requiredPositiveInt(index: Int, name: String): Int = checkedNonNegativeInt(getLong(index), name).also { check(it > 0) { "$name doit être positif" } }
+    private fun android.database.Cursor.finiteNonNegativeDouble(index: Int, name: String, strictlyPositive: Boolean = false): Double = getDouble(index).also { check(it.isFinite() && if (strictlyPositive) it > 0.0 else it >= 0.0) { "$name invalide" } }
+    private fun android.database.Cursor.optionalFinitePositiveDouble(index: Int, name: String): Double? = if (isNull(index)) null else finiteNonNegativeDouble(index, name, true)
+    private fun checkedNonNegativeInt(value: Long, name: String): Int { check(value in 0..Int.MAX_VALUE.toLong()) { "$name hors plage" }; return value.toInt() }
+
+    private companion object {
+        const val MAX_OCCURRENCE_PAGE_SIZE = 32
+        const val MAX_SET_PAGE_SIZE = 64
+    }
+
     /**
      * Return one newest explicit measured max per movement. Equipment is
      * presentation context only and never participates in max identity.

android/app/src/main/java/com/labfytools/trainlog/data/TrainlogTimestamp.kt

@@ -0,0 +1,106 @@
+package com.labfytools.trainlog.data
+
+internal data class TrainlogTimestampKey(
+    val utcSecond: Long,
+    val fraction: String,
+) : Comparable<TrainlogTimestampKey> {
+    override fun compareTo(other: TrainlogTimestampKey): Int {
+        utcSecond.compareTo(other.utcSecond).takeIf { it != 0 }?.let { return it }
+        val count = maxOf(fraction.length, other.fraction.length)
+        for (index in 0 until count) {
+            val left = fraction.getOrElse(index) { '0' }
+            val right = other.fraction.getOrElse(index) { '0' }
+            if (left != right) return left.compareTo(right)
+        }
+        return 0
+    }
+}
+
+internal object TrainlogTimestamp {
+    /**
+     * CONTRACT: parse the frozen ASCII timestamp grammar without inheriting a
+     * platform ISO parser's syntax, offset bounds, or fractional precision.
+     */
+    fun parse(value: String): TrainlogTimestampKey? {
+        if (value.length < 17 || value.any { it.code > 0x7f }) return null
+        val year = digits(value, 0, 4) ?: return null
+        val month = digits(value, 5, 2) ?: return null
+        val day = digits(value, 8, 2) ?: return null
+        val hour = digits(value, 11, 2) ?: return null
+        val minute = digits(value, 14, 2) ?: return null
+        if (value.getOrNull(4) != '-' || value.getOrNull(7) != '-' ||
+            value.getOrNull(10) !in listOf('T', 't') || value.getOrNull(13) != ':' ||
+            year !in 1..9999 || month !in 1..12 || day !in 1..daysInMonth(year, month) ||
+            hour !in 0..23 || minute !in 0..59
+        ) return null
+
+        var at = 16
+        var second = 0
+        var fraction = ""
+        if (value.getOrNull(at) == ':') {
+            second = digits(value, at + 1, 2) ?: return null
+            if (second !in 0..59) return null
+            at += 3
+            if (value.getOrNull(at) == '.') {
+                val start = ++at
+                while (value.getOrNull(at)?.isAsciiDigit() == true) at++
+                if (at == start) return null
+                fraction = value.substring(start, at)
+            }
+        }
+
+        var offsetSeconds = 0L
+        if (at + 1 == value.length && value[at] in listOf('Z', 'z')) {
+            // UTC designator.
+        } else {
+            if (at + 6 != value.length || value.getOrNull(at) !in listOf('+', '-') ||
+                value.getOrNull(at + 3) != ':'
+            ) return null
+            val offsetHour = digits(value, at + 1, 2) ?: return null
+            val offsetMinute = digits(value, at + 4, 2) ?: return null
+            if (offsetHour !in 0..23 || offsetMinute !in 0..59) return null
+            offsetSeconds = (offsetHour * 3600L + offsetMinute * 60L) *
+                if (value[at] == '+') 1L else -1L
+        }
+        val localSecond = dayNumber(year, month, day) * 86400L +
+            hour * 3600L + minute * 60L + second
+        return TrainlogTimestampKey(localSecond - offsetSeconds, fraction)
+    }
+
+    /** Bytewise identity order shared with C's unsigned memcmp semantics. */
+    fun compareIds(left: String, right: String): Int {
+        val leftBytes = left.toByteArray(Charsets.UTF_8)
+        val rightBytes = right.toByteArray(Charsets.UTF_8)
+        for (index in 0 until minOf(leftBytes.size, rightBytes.size)) {
+            val result = (leftBytes[index].toInt() and 0xff).compareTo(rightBytes[index].toInt() and 0xff)
+            if (result != 0) return result
+        }
+        return leftBytes.size.compareTo(rightBytes.size)
+    }
+
+    private fun digits(value: String, start: Int, count: Int): Int? {
+        if (start < 0 || start + count > value.length) return null
+        var result = 0
+        repeat(count) { offset ->
+            val character = value[start + offset]
+            if (!character.isAsciiDigit()) return null
+            result = result * 10 + (character - '0')
+        }
+        return result
+    }
+
+    private fun Char.isAsciiDigit(): Boolean = this in '0'..'9'
+    private fun leap(year: Int): Boolean = year % 4 == 0 && (year % 100 != 0 || year % 400 == 0)
+    private fun daysInMonth(year: Int, month: Int): Int = when (month) {
+        2 -> if (leap(year)) 29 else 28
+        4, 6, 9, 11 -> 30
+        else -> 31
+    }
+
+    private fun dayNumber(year: Int, month: Int, day: Int): Long {
+        val prior = year - 1
+        var days = prior * 365L + prior / 4 - prior / 100 + prior / 400
+        for (current in 1 until month) days += daysInMonth(year, current)
+        return days + day - 1
+    }
+}

android/app/src/main/java/com/labfytools/trainlog/ui/ExerciseScreen.kt

@@ -19,6 +19,9 @@ import com.labfytools.trainlog.data.CreateExerciseResult
 import com.labfytools.trainlog.data.BodyZone
 import com.labfytools.trainlog.data.BodyZoneKind
 import com.labfytools.trainlog.data.EditExerciseResult
+import com.labfytools.trainlog.data.ExerciseKnowledge
+import com.labfytools.trainlog.data.ExerciseKnowledgeStatus
+import com.labfytools.trainlog.data.KnowledgeConfidence
 import com.labfytools.trainlog.data.TrainlogRepository
 import com.labfytools.trainlog.model.ExerciseDataFields
 import com.labfytools.trainlog.model.ExerciseEditInput
@@ -73,6 +76,7 @@ fun ExerciseScreen(
     var searchQuery by remember { mutableStateOf("") }
     var filterZoneId by remember { mutableStateOf<String?>(null) }
     var unclassifiedFilter by remember { mutableStateOf(false) }
+    var expandedKnowledgeIds by remember { mutableStateOf(emptySet<String>()) }
     val zones = repository.listBodyZones()
 
     var message by
@@ -513,6 +517,22 @@ fun ExerciseScreen(
                         accent = colors.accent,
                         onClick = { startEditing(exercise) },
                     )
+                    repository.getExerciseKnowledge(exercise.exerciseId)?.let { knowledge ->
+                        val expanded = exercise.exerciseId in expandedKnowledgeIds
+                        TrainlogAction(
+                            label = if (expanded) "− Connaissances" else "+ Connaissances",
+                            description = knowledgeSummary(knowledge),
+                            accent = colors.muted,
+                            onClick = {
+                                expandedKnowledgeIds = if (expanded) {
+                                    expandedKnowledgeIds - exercise.exerciseId
+                                } else {
+                                    expandedKnowledgeIds + exercise.exerciseId
+                                }
+                            },
+                        )
+                        if (expanded) KnowledgePanel(repository, knowledge)
+                    }
                 }
             }
         }
@@ -532,6 +552,59 @@ fun ExerciseScreen(
     }
 }
 
+@Composable
+private fun KnowledgePanel(repository: TrainlogRepository, knowledge: ExerciseKnowledge) {
+    val colors = LocalTrainlogColors.current
+    /* WHY: conditional content is opened only under an explicit uncertainty
+     * label; it cannot be mistaken for an ordinary resolved classification. */
+    val interpretation = when (knowledge.resolutionStatus) {
+        ExerciseKnowledgeStatus.RESOLVED_FAMILY_VARIANT_LIMITED -> knowledge.interpretation
+        ExerciseKnowledgeStatus.CONDITIONAL -> knowledge.conditionalInterpretation
+        ExerciseKnowledgeStatus.UNRESOLVED -> null
+    }
+    if (interpretation == null) {
+        TrainlogInfo("Classification scientifique non résolue.", colors.warning)
+        return
+    }
+    if (knowledge.resolutionStatus == ExerciseKnowledgeStatus.CONDITIONAL) {
+        TrainlogInfo(
+            "Interprétation conditionnelle · à confirmer : ${interpretation.requiredConfirmation.orEmpty()}",
+            colors.warning,
+        )
+    }
+    val patterns = interpretation.patternIds.mapNotNull(repository::getMovementPatternKnowledge)
+    val primary = interpretation.primaryMuscleIds.mapNotNull(repository::getMuscleKnowledge)
+    val secondary = interpretation.secondaryMuscleIds.mapNotNull(repository::getMuscleKnowledge)
+    val primaryZone = repository.bodyZone(interpretation.primaryZoneId)?.displayName
+        ?: interpretation.primaryZoneId
+    val secondaryZones = interpretation.secondaryZoneIds.map { repository.bodyZone(it)?.displayName ?: it }
+    val runtimeEquipment = repository.listEquipment().associateBy { it.equipmentId }
+    val equipment = knowledge.equipmentIds.map { runtimeEquipment[it]?.displayName ?: it }
+    TrainlogInfo("Mouvement : ${patterns.joinToString { it.displayNameFr }.ifEmpty { "Non classé" }}")
+    TrainlogInfo("Muscles principaux : ${primary.joinToString { it.displayNameFr }.ifEmpty { "Non classés" }}")
+    TrainlogInfo("Muscles secondaires : ${secondary.joinToString { it.displayNameFr }.ifEmpty { "Aucun établi" }}")
+    TrainlogInfo(
+        "Zones scientifiques : $primaryZone" +
+            if (secondaryZones.isEmpty()) "" else " · secondaires : ${secondaryZones.joinToString()}",
+    )
+    TrainlogInfo("Équipement compatible : ${equipment.joinToString().ifEmpty { "Non établi" }}")
+    TrainlogInfo("Confiance : ${confidenceLabel(interpretation.confidence)}", colors.muted)
+}
+
+private fun knowledgeSummary(knowledge: ExerciseKnowledge): String = when (knowledge.resolutionStatus) {
+    ExerciseKnowledgeStatus.RESOLVED_FAMILY_VARIANT_LIMITED ->
+        "Classification scientifique · confiance ${confidenceLabel(knowledge.confidence)}"
+    ExerciseKnowledgeStatus.CONDITIONAL ->
+        "Classification conditionnelle · incertitude explicite"
+    ExerciseKnowledgeStatus.UNRESOLVED -> "Classification scientifique non résolue"
+}
+
+private fun confidenceLabel(confidence: KnowledgeConfidence): String = when (confidence) {
+    KnowledgeConfidence.HIGH -> "élevée"
+    KnowledgeConfidence.MODERATE -> "modérée"
+    KnowledgeConfidence.UNCERTAIN -> "incertaine"
+}
+
 @Composable
 private fun BodyZoneChoices(
     zones: List<BodyZone>,

android/app/src/test/java/com/labfytools/trainlog/data/TrainingExerciseContextTest.kt

@@ -0,0 +1,206 @@
+package com.labfytools.trainlog.data
+
+import android.content.Context
+import android.database.sqlite.SQLiteDatabase
+import androidx.test.core.app.ApplicationProvider
+import com.labfytools.trainlog.model.ExerciseEditInput
+import org.junit.After
+import org.junit.Assert.assertEquals
+import org.junit.Assert.assertFalse
+import org.junit.Assert.assertNull
+import org.junit.Assert.assertTrue
+import org.junit.Test
+import org.junit.runner.RunWith
+import org.robolectric.RobolectricTestRunner
+import org.robolectric.annotation.Config
+import java.util.UUID
+
+@RunWith(RobolectricTestRunner::class)
+@Config(sdk = [35])
+class TrainingExerciseContextTest {
+    private val context: Context = ApplicationProvider.getApplicationContext()
+    private val databaseName = "knowledge-context-${UUID.randomUUID()}.db"
+    private val repository = TrainlogRepository(context, databaseName)
+
+    @After fun close() { repository.close(); context.deleteDatabase(databaseName) }
+
+    @Test
+    fun composesExactIdentityZonesEquipmentMaxAndBoundedOccurrencePages() {
+        seedFixture()
+        val exerciseId = LEG_PRESS_ID
+        val first = repository.getTrainingExerciseContext(exerciseId, occurrenceLimit = 2, setPreviewLimit = 2)!!
+        assertEquals("Nom modifiable", first.exercise.name)
+        assertEquals(listOf("thighs", "glutes"), first.persistedDirectZoneIds)
+        assertTrue("lower_body" in first.persistedZoneIdsWithAncestors)
+        assertEquals("thighs", repository.getScientificBodyZoneMapping(exerciseId)?.primaryZoneId)
+        assertTrue(first.compatibleEquipment.any { it.equipmentId == "leg_press" })
+
+        val maximum = first.latestExplicitMax!!
+        assertEquals(140.0, maximum.maxWeightKg, 0.0)
+        assertEquals(EquipmentLoadSemantics.ASSISTANCE, maximum.loadSemantics)
+        assertFalse(first.recentPerformance.occurrences.any { occurrence ->
+            occurrence.setPreview?.sets.orEmpty().any { it.weightKg == 999.0 } && occurrence.sessionType == com.labfytools.trainlog.model.SessionType.MAX_TEST
+        })
+        assertEquals(2, first.recentPerformance.occurrences.size)
+        val cursor = first.recentPerformance.nextCursor!!
+        val second = repository.listExerciseOccurrences(exerciseId, 2, cursor, 2)
+        assertEquals(2, second.occurrences.size)
+        assertTrue(first.recentPerformance.occurrences.map { it.entryId }.toSet().intersect(second.occurrences.map { it.entryId }.toSet()).isEmpty())
+
+        val assistanceOccurrence = (first.recentPerformance.occurrences + second.occurrences).first { it.entryId == "entry_a" }
+        assertEquals(EquipmentLoadSemantics.ASSISTANCE, assistanceOccurrence.loadSemantics)
+        assertEquals(listOf(null, 0.0), assistanceOccurrence.setPreview!!.sets.map { it.weightKg })
+        val setPage2 = repository.listExerciseOccurrenceSets(exerciseId, "entry_a", 2, assistanceOccurrence.setPreview.nextPosition)
+        assertEquals(listOf(22.5), setPage2.sets.map { it.weightKg })
+        assertNull(setPage2.nextPosition)
+
+        assertTrue(repository.getExerciseKnowledge(exerciseId) != null)
+        assertNull(repository.getTrainingExerciseContext("ex_00000000-0000-4000-8000-000000000000"))
+    }
+
+    @Test
+    fun continuousOccurrenceHasNoArtificialSetsAndInvalidPagingFails() {
+        seedFixture()
+        val continuous = repository.getTrainingExerciseContext(CONTINUOUS_ID)!!.recentPerformance.occurrences.single()
+        assertNull(continuous.setPreview)
+        assertEquals(1800, continuous.continuousDurationSeconds)
+        assertEquals(8.0, continuous.speedKmh!!, 0.0)
+        assertTrue(runCatching { repository.listExerciseOccurrences(LEG_PRESS_ID, 0) }.isFailure)
+        assertTrue(runCatching { repository.listExerciseOccurrences(LEG_PRESS_ID, 2, ExerciseOccurrenceCursor("bad", "s", "e")) }.isFailure)
+        assertTrue(runCatching { repository.listExerciseOccurrenceSets(CONTINUOUS_ID, "entry_cont", 2) }.isFailure)
+    }
+
+    @Test
+    fun occurrencesAndExplicitMaxOrderByInstantAcrossOffsetsWithStablePagingTies() {
+        seedFixture()
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            listOf(
+                arrayOf<Any?>("offset_local_older", "2026-02-01T10:00:00+02:00", "training", "entry_offset_old"),
+                arrayOf<Any?>("offset_utc_newer", "2026-02-01T09:00:00Z", "training", "entry_offset_new"),
+                arrayOf<Any?>("tie_a", "2026-02-02T10:00:00+02:00", "training", "entry_tie_a"),
+                arrayOf<Any?>("tie_z", "2026-02-02T08:00:00Z", "training", "entry_tie_z"),
+                arrayOf<Any?>("max_local_older", "2025-03-01T10:00:00+02:00", "max_test", "entry_max_old"),
+                arrayOf<Any?>("max_utc_newer", "2025-03-01T09:00:00Z", "max_test", "entry_max_new"),
+            ).forEach { row ->
+                db.execSQL("INSERT INTO sessions(session_id,started_at,session_type) VALUES(?,?,?);", row.copyOfRange(0, 3))
+                db.execSQL("INSERT INTO session_exercises(session_row_id,exercise_row_id,position,recording_mode,tracking_mode,data_fields,entry_id) SELECT s.id,e.id,0,'sets','reps',0,? FROM sessions s JOIN exercises e ON e.exercise_id=? WHERE s.session_id=?;", arrayOf(row[3], LEG_PRESS_ID, row[0]))
+            }
+            db.execSQL("INSERT INTO max_results(session_exercise_row_id,max_weight_kg) SELECT id,150.0 FROM session_exercises WHERE entry_id='entry_max_old';")
+            db.execSQL("INSERT INTO max_results(session_exercise_row_id,max_weight_kg) SELECT id,160.0 FROM session_exercises WHERE entry_id='entry_max_new';")
+        }
+
+        val tieFirst = repository.listExerciseOccurrences(LEG_PRESS_ID, 1)
+        assertEquals("tie_z", tieFirst.occurrences.single().sessionId)
+        val tieSecond = repository.listExerciseOccurrences(LEG_PRESS_ID, 1, tieFirst.nextCursor)
+        assertEquals("tie_a", tieSecond.occurrences.single().sessionId)
+        val following = repository.listExerciseOccurrences(LEG_PRESS_ID, 2, tieSecond.nextCursor)
+        assertEquals(listOf("offset_utc_newer", "offset_local_older"), following.occurrences.map { it.sessionId })
+
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            db.execSQL("UPDATE sessions SET started_at='2026-03-01T10:00:00+02:00' WHERE session_id='max_local_older';")
+            db.execSQL("UPDATE sessions SET started_at='2026-03-01T09:00:00Z' WHERE session_id='max_utc_newer';")
+        }
+        val maximum = repository.getTrainingExerciseContext(LEG_PRESS_ID)!!.latestExplicitMax!!
+        assertEquals("max_utc_newer", maximum.sessionId)
+        assertEquals(160.0, maximum.maxWeightKg, 0.0)
+    }
+
+    @Test
+    fun exceptionalTimestampSpellingsRoundTripWithExactFractionsAndCorruptionFails() {
+        seedFixture()
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            listOf(
+                arrayOf("omit", "2030-01-05T10:00+15:00", "entry_omit"),
+                arrayOf("frac_low", "2030-01-04T10:00:00.12345678901234567890Z", "entry_frac_low"),
+                arrayOf("frac_high", "2030-01-04T10:00:00.12345678901234567891Z", "entry_frac_high"),
+                arrayOf("tie_a", "2030-01-04T12:00:00+02:00", "entry_tie_a2"),
+                arrayOf("tie_z", "2030-01-04T10:00:00-00:00", "entry_tie_z2"),
+                arrayOf("lower", "2030-01-03t10:00:00z", "entry_lower"),
+                arrayOf("high_offset", "2030-01-04T09:00:00+23:59", "entry_high_offset"),
+            ).forEach { row ->
+                db.execSQL("INSERT INTO sessions(session_id,started_at,session_type) VALUES(?,?,'training');", arrayOf(row[0], row[1]))
+                db.execSQL("INSERT INTO session_exercises(session_row_id,exercise_row_id,position,recording_mode,tracking_mode,data_fields,entry_id) SELECT s.id,e.id,0,'sets','reps',0,? FROM sessions s JOIN exercises e ON e.exercise_id=? WHERE s.session_id=?;", arrayOf(row[2], LEG_PRESS_ID, row[0]))
+            }
+            listOf(
+                arrayOf<Any?>("max_offset", "2031-01-02T10:00:00+15:00", "entry_max_offset", 151.0),
+                arrayOf<Any?>("max_true", "2031-01-01T20:00:00Z", "entry_max_true", 152.0),
+            ).forEach { row ->
+                db.execSQL("INSERT INTO sessions(session_id,started_at,session_type) VALUES(?,?,'max_test');", arrayOf(row[0], row[1]))
+                db.execSQL("INSERT INTO session_exercises(session_row_id,exercise_row_id,position,recording_mode,tracking_mode,data_fields,entry_id) SELECT s.id,e.id,0,'sets','reps',0,? FROM sessions s JOIN exercises e ON e.exercise_id=? WHERE s.session_id=?;", arrayOf(row[2], LEG_PRESS_ID, row[0]))
+                db.execSQL("INSERT INTO max_results(session_exercise_row_id,max_weight_kg) SELECT id,? FROM session_exercises WHERE entry_id=?;", arrayOf(row[3], row[2]))
+            }
+        }
+
+        val expected = listOf("max_true", "max_offset", "omit", "frac_high", "frac_low", "tie_z", "tie_a", "lower", "high_offset")
+        var cursor: ExerciseOccurrenceCursor? = null
+        val seen = mutableSetOf<String>()
+        expected.forEach { sessionId ->
+            val page = repository.listExerciseOccurrences(LEG_PRESS_ID, 1, cursor)
+            assertEquals(sessionId, page.occurrences.single().sessionId)
+            assertTrue(seen.add(page.occurrences.single().entryId))
+            assertEquals(page.occurrences.single().startedAt, page.nextCursor!!.startedAt)
+            cursor = page.nextCursor
+        }
+        while (cursor != null) {
+            val page = repository.listExerciseOccurrences(LEG_PRESS_ID, 1, cursor)
+            page.occurrences.forEach { assertTrue(seen.add(it.entryId)) }
+            cursor = page.nextCursor
+        }
+        val maximum = repository.getTrainingExerciseContext(LEG_PRESS_ID)!!.latestExplicitMax!!
+        assertEquals("max_true", maximum.sessionId)
+        assertEquals(152.0, maximum.maxWeightKg, 0.0)
+
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            db.execSQL("UPDATE sessions SET started_at='2030-01-03 10:00:00Z' WHERE session_id='lower';")
+        }
+        assertTrue(runCatching { repository.listExerciseOccurrences(LEG_PRESS_ID, 1) }.isFailure)
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            db.execSQL("UPDATE sessions SET started_at='2030-01-03T10:00:00Z' WHERE session_id='lower';")
+            db.execSQL("UPDATE sessions SET started_at='bad-max' WHERE session_id='max_offset';")
+        }
+        assertTrue(runCatching { repository.getTrainingExerciseContext(LEG_PRESS_ID) }.isFailure)
+    }
+
+    private fun seedFixture() {
+        repository.listExercises()
+        SQLiteDatabase.openDatabase(context.getDatabasePath(databaseName).path, null, SQLiteDatabase.OPEN_READWRITE).use { db ->
+            db.execSQL("PRAGMA foreign_keys=ON;")
+            db.execSQL("INSERT INTO exercises(exercise_id,name,normalized_name,recording_mode,tracking_mode,data_fields) VALUES(?,?,?,?,?,?);", arrayOf<Any?>(LEG_PRESS_ID, "Nom original", "nom original", "sets", "reps", 0))
+            db.execSQL("INSERT INTO exercises(exercise_id,name,normalized_name,recording_mode,tracking_mode,data_fields) VALUES(?,?,?,?,?,?);", arrayOf<Any?>(CONTINUOUS_ID, "Marche test", "marche test", "continuous", "duration", 3))
+            db.execSQL("INSERT INTO exercise_body_zones(exercise_row_id,zone_id,role) SELECT id,'thighs','primary' FROM exercises WHERE exercise_id=?;", arrayOf(LEG_PRESS_ID))
+            db.execSQL("INSERT INTO exercise_body_zones(exercise_row_id,zone_id,role) SELECT id,'glutes','secondary' FROM exercises WHERE exercise_id=?;", arrayOf(LEG_PRESS_ID))
+            listOf(
+                arrayOf<Any?>("session_latest", "2026-01-04T10:00:00+00:00", "training"),
+                arrayOf<Any?>("session_tied", "2026-01-03T10:00:00+00:00", "training"),
+                arrayOf<Any?>("session_max", "2026-01-02T10:00:00+00:00", "max_test"),
+                arrayOf<Any?>("session_old", "2026-01-01T10:00:00+00:00", "training"),
+                arrayOf<Any?>("session_cont", "2026-01-05T10:00:00+00:00", "training"),
+            ).forEach { db.execSQL("INSERT INTO sessions(session_id,started_at,session_type) VALUES(?,?,?);", it) }
+            fun occurrence(session: String, entry: String, position: Int, equipment: String? = null): Long {
+                db.execSQL(
+                    "INSERT INTO session_exercises(session_row_id,exercise_row_id,position,recording_mode,tracking_mode,data_fields,equipment_row_id,entry_id) SELECT s.id,e.id,?,'sets','reps',0,eq.id,? FROM sessions s JOIN exercises e ON e.exercise_id=? LEFT JOIN equipment eq ON eq.equipment_id=? WHERE s.session_id=?;",
+                    arrayOf<Any?>(position, entry, LEG_PRESS_ID, equipment, session),
+                )
+                return db.rawQuery("SELECT id FROM session_exercises WHERE entry_id=?;", arrayOf(entry)).use { it.moveToFirst(); it.getLong(0) }
+            }
+            val latest = occurrence("session_latest", "entry_latest", 0, "leg_press")
+            db.execSQL("INSERT INTO performed_sets(session_exercise_row_id,position,reps,weight_kg) VALUES(?,0,3,999.0);", arrayOf(latest))
+            val a = occurrence("session_tied", "entry_a", 0, "assisted_dip_chin_machine")
+            db.execSQL("INSERT INTO performed_sets(session_exercise_row_id,position,reps,weight_kg) VALUES(?,0,8,NULL),(?,1,7,0.0),(?,2,6,22.5);", arrayOf(a, a, a))
+            occurrence("session_tied", "entry_b", 1, "plate_loaded_leg_press")
+            val max = occurrence("session_max", "entry_max", 0, "assisted_dip_chin_machine")
+            db.execSQL("INSERT INTO max_results(session_exercise_row_id,max_weight_kg) VALUES(?,140.0);", arrayOf(max))
+            occurrence("session_old", "entry_old", 0)
+            db.execSQL("INSERT INTO session_exercises(session_row_id,exercise_row_id,position,recording_mode,tracking_mode,data_fields,equipment_row_id,entry_id) SELECT s.id,e.id,0,'continuous','duration',3,eq.id,'entry_cont' FROM sessions s JOIN exercises e ON e.exercise_id=? LEFT JOIN equipment eq ON eq.equipment_id='treadmill' WHERE s.session_id='session_cont';", arrayOf(CONTINUOUS_ID))
+            db.execSQL("INSERT INTO continuous_activity(session_exercise_row_id,duration_seconds,speed_kmh,distance_km) SELECT id,1800,8.0,4.0 FROM session_exercises WHERE entry_id='entry_cont';")
+        }
+        val profile = repository.listExercises().first { it.exerciseId == LEG_PRESS_ID }
+        val result = repository.editExercise(ExerciseEditInput(profile.exerciseId, "Nom modifiable", profile.recordingMode, profile.trackingMode, profile.dataFields, profile.primaryZoneId, profile.secondaryZoneIds))
+        assertTrue(result is EditExerciseResult.Saved)
+    }
+
+    companion object {
+        private const val LEG_PRESS_ID = "ex_b432623f-bfe9-4daf-a653-60ec7fdffbde"
+        private const val CONTINUOUS_ID = "ex_00000000-0000-4000-8000-000000000001"
+    }
+}

android/app/src/test/java/com/labfytools/trainlog/data/TrainingKnowledgeCatalogTest.kt

@@ -0,0 +1,182 @@
+package com.labfytools.trainlog.data
+
+import android.content.Context
+import androidx.test.core.app.ApplicationProvider
+import org.json.JSONObject
+import org.junit.Assert.assertEquals
+import org.junit.Assert.assertFalse
+import org.junit.Assert.assertNotNull
+import org.junit.Assert.assertNull
+import org.junit.Assert.assertTrue
+import org.junit.Test
+import org.junit.runner.RunWith
+import org.robolectric.RobolectricTestRunner
+import org.robolectric.annotation.Config
+
+@RunWith(RobolectricTestRunner::class)
+@Config(sdk = [35])
+class TrainingKnowledgeCatalogTest {
+    private val context: Context = ApplicationProvider.getApplicationContext()
+    private val names = listOf(
+        "science-references-v1.json", "muscles-v1.json", "joint-actions-v1.json",
+        "movement-patterns-v1.json", "exercise-knowledge-v1.json", "equipment-knowledge-v1.json",
+    )
+
+    @Test
+    fun loadsAllStructuredCatalogsAndRepresentativeMappings() {
+        val catalog = TrainingKnowledgeCatalog.load(context)
+        assertEquals(listOf(23, 53, 35, 26, 23, 41), listOf(catalog.references.size, catalog.muscles.size, catalog.jointActions.size, catalog.movementPatterns.size, catalog.exercises.size, catalog.equipment.size))
+        val legPress = catalog.getExerciseKnowledge("ex_b432623f-bfe9-4daf-a653-60ec7fdffbde")
+        assertEquals("knee_dominant", legPress?.interpretation?.patternIds?.single())
+        assertTrue(catalog.listExercisesByMovementPattern("knee_dominant").any { it.exerciseId == legPress?.exerciseId })
+        assertTrue(catalog.listExercisesByMuscle("quadriceps", MuscleRole.PRIMARY).any { it.exerciseId == legPress?.exerciseId })
+        assertEquals("thighs", catalog.getScientificBodyZoneMapping(legPress!!.exerciseId)?.primaryZoneId)
+        assertEquals(BodyZoneAuditStatus.CONFIRMED, legPress.bodyZoneAudit.status)
+        assertEquals("thighs", legPress.bodyZoneAudit.existingPrimaryZoneId)
+        assertTrue(legPress.bodyZoneAudit.sourceRefs.isNotEmpty())
+
+        val multifunction = catalog.getEquipmentKnowledge("rear_delt_pec_fly")
+        assertEquals(2, multifunction?.capabilities?.size)
+        assertTrue(catalog.listCompatibleExercises("rear_delt_pec_fly").any { it.exerciseId == "ex_4cd2433e-80b1-478a-b8df-73fc6ef80962" })
+        assertNotNull(catalog.getMuscle("deltoid_posterior"))
+        assertNotNull(catalog.getJointAction("shoulder_horizontal_abduction"))
+        assertNotNull(catalog.getReference(legPress.sourceRefs.first()))
+    }
+
+    @Test
+    fun conditionalAndUnknownRecordsCannotLeakIntoOrdinaryQueries() {
+        val catalog = TrainingKnowledgeCatalog.load(context)
+        val chestPress = "ex_8552dd77-fcd7-4f06-a1cc-d956eb1009af"
+        assertEquals(ExerciseKnowledgeStatus.CONDITIONAL, catalog.getConditionalExerciseKnowledge(chestPress)?.resolutionStatus)
+        assertNull(catalog.getScientificBodyZoneMapping(chestPress))
+        assertFalse(catalog.queryExercises(KnowledgeExerciseFilters()).any { it.exerciseId == chestPress })
+        assertNull(catalog.getExerciseKnowledge("ex_00000000-0000-4000-8000-000000000000"))
+    }
+
+    @Test
+    fun acceptsAdditiveNonRuntimeKnowledgeRecord() {
+        val files = assetFiles()
+        val root = JSONObject(files.getValue("science-references-v1.json"))
+        root.getJSONArray("references").put(
+            JSONObject()
+                .put("ref_id", "zz_additive_reference")
+                .put("title", "Additive reference fixture")
+                .put("authors_or_organization", "Trainlog test")
+                .put("year", 2026)
+                .put("type", "test_fixture")
+                .put("url", "https://example.invalid/additive-reference")
+                .put("topics", org.json.JSONArray().put("validation"))
+                .put("notes", "Valid additive metadata record.")
+                .put("limitations", "Test fixture only.")
+                .put("doi", JSONObject.NULL)
+                .put("pmid", JSONObject.NULL)
+                .put("accessed_on", "2026-09-09"),
+        )
+        files["science-references-v1.json"] = root.toString()
+
+        val catalog = TrainingKnowledgeCatalog.load(files::getValue, BodyZoneCatalog.load(context))
+        assertNotNull(catalog.getReference("zz_additive_reference"))
+    }
+
+    @Test
+    fun rejectsVersionEnumsDanglingReferencesAndDuplicateKeys() {
+        assertInvalid { files -> files["muscles-v1.json"] = files.getValue("muscles-v1.json").replaceFirst("\"version\": 1", "\"version\": 2") }
+        assertInvalid { files -> files["muscles-v1.json"] = files.getValue("muscles-v1.json").replaceFirst("\"entity_type\": \"muscle\"", "\"entity_type\": \"organ\"") }
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("exercise-knowledge-v1.json")); root.getJSONArray("exercises").getJSONObject(0).getJSONArray("source_refs").put("missing_ref")
+            files["exercise-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalid { files -> files["science-references-v1.json"] = files.getValue("science-references-v1.json").replaceFirst("{", "{\"version\":1,") }
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("exercise-knowledge-v1.json"))
+            val legPress = root.getJSONArray("exercises").let { rows -> (0 until rows.length()).map(rows::getJSONObject).first { it.getString("exercise_id") == "ex_b432623f-bfe9-4daf-a653-60ec7fdffbde" } }
+            val equipment = legPress.getJSONArray("equipment_ids"); val first = equipment.getString(0)
+            equipment.put(0, equipment.getString(1)); equipment.put(1, first)
+            files["exercise-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("exercise-knowledge-v1.json")); val interpretation = root.getJSONArray("exercises").getJSONObject(0).getJSONObject("interpretation")
+            interpretation.getJSONArray("secondary_muscle_ids").put(interpretation.getJSONArray("primary_muscle_ids").getString(0))
+            files["exercise-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("equipment-knowledge-v1.json")); val equipment = root.getJSONArray("equipment").getJSONObject(0)
+            equipment.put("confidence", "high"); equipment.put("evidence_type", "manufacturer_statement")
+            files["equipment-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalidAudit { it.getJSONArray("source_refs").put("missing_ref") }
+        assertInvalidAudit { it.put("existing_primary_zone_id", "missing_zone") }
+        assertInvalidAudit { it.put("status", "compatible") }
+    }
+
+    @Test
+    fun rejectsInvalidRuntimeIdentitySyntax() {
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("exercise-knowledge-v1.json"))
+            val row = root.getJSONArray("exercises").getJSONObject(0)
+            row.put("exercise_id", row.getString("exercise_id").dropLast(1) + "g")
+            files["exercise-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("equipment-knowledge-v1.json"))
+            val rows = root.getJSONArray("equipment")
+            val row = rows.getJSONObject(0)
+            val oldId = row.getString("equipment_id")
+            val invalidId = "$oldId-"
+            row.put("equipment_id", invalidId)
+            val exercises = JSONObject(files.getValue("exercise-knowledge-v1.json"))
+            val exerciseRows = exercises.getJSONArray("exercises")
+            for (index in 0 until exerciseRows.length()) {
+                val ids = exerciseRows.getJSONObject(index).getJSONArray("equipment_ids")
+                for (idIndex in 0 until ids.length()) if (ids.getString(idIndex) == oldId) ids.put(idIndex, invalidId)
+            }
+            files["equipment-knowledge-v1.json"] = root.toString()
+            files["exercise-knowledge-v1.json"] = exercises.toString()
+        }
+    }
+
+    @Test
+    fun rejectsReverseCompatibilityHighEvidenceAndResolvedAuditGaps() {
+        assertInvalid { files ->
+            val root = JSONObject(files.getValue("exercise-knowledge-v1.json"))
+            val rows = root.getJSONArray("exercises")
+            val row = (0 until rows.length()).map(rows::getJSONObject).first { it.getJSONArray("equipment_ids").length() > 0 }
+            row.put("equipment_ids", org.json.JSONArray())
+            files["exercise-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalid { files ->
+            val references = JSONObject(files.getValue("science-references-v1.json")).getJSONArray("references")
+            val nonScientificRef = (0 until references.length()).map(references::getJSONObject)
+                .first { it.getString("type") !in setOf("established_anatomy", "emg_evidence", "intervention_evidence") }
+                .getString("ref_id")
+            val root = JSONObject(files.getValue("equipment-knowledge-v1.json"))
+            val rows = root.getJSONArray("equipment")
+            val row = rows.getJSONObject(0)
+            row.put("confidence", "high")
+            row.put("source_refs", org.json.JSONArray().put(nonScientificRef))
+            row.put("evidence_type", "mixed_evidence")
+            files["equipment-knowledge-v1.json"] = root.toString()
+        }
+        assertInvalidAudit { audit ->
+            audit.put("source_refs", org.json.JSONArray())
+            audit.put("status", "confirmed")
+        }
+    }
+
+    private fun assertInvalid(mutate: (MutableMap<String, String>) -> Unit) {
+        val files = assetFiles()
+        mutate(files)
+        val failed = runCatching { TrainingKnowledgeCatalog.load(files::getValue, BodyZoneCatalog.load(context)) }.isFailure
+        assertTrue("catalogue corrompu accepté", failed)
+    }
+
+    private fun assetFiles(): MutableMap<String, String> = names.associateWith { name ->
+        context.assets.open(name).bufferedReader().use { it.readText() }
+    }.toMutableMap()
+
+    private fun assertInvalidAudit(mutate: (JSONObject) -> Unit) = assertInvalid { files ->
+        val root = JSONObject(files.getValue("exercise-knowledge-v1.json"))
+        mutate(root.getJSONArray("exercises").getJSONObject(0).getJSONObject("body_zone_audit"))
+        files["exercise-knowledge-v1.json"] = root.toString()
+    }
+}

android/app/src/test/java/com/labfytools/trainlog/data/TrainlogTimestampTest.kt

@@ -0,0 +1,33 @@
+package com.labfytools.trainlog.data
+
+import org.junit.Assert.assertEquals
+import org.junit.Assert.assertNotNull
+import org.junit.Assert.assertNull
+import org.junit.Assert.assertTrue
+import org.junit.Test
+
+class TrainlogTimestampTest {
+    @Test fun acceptsSettledGrammarAndComparesEveryFractionDigit() {
+        listOf(
+            "0001-01-01T00:00Z", "2000-02-29t23:59:59.1z",
+            "2026-09-05T18:34+23:59", "2026-09-05T18:34:12-00:00",
+            "9999-12-31T23:59:59.9-23:59",
+        ).forEach { assertNotNull(it, TrainlogTimestamp.parse(it)) }
+        val low = TrainlogTimestamp.parse("2026-01-01T00:00:00.12345678901234567890Z")!!
+        val high = TrainlogTimestamp.parse("2026-01-01T00:00:00.12345678901234567891Z")!!
+        val equal = TrainlogTimestamp.parse("2026-01-01T00:00:00.123456789012345678900Z")!!
+        assertTrue(low < high)
+        assertEquals(0, low.compareTo(equal))
+    }
+
+    @Test fun rejectsPlatformOnlyAndOutOfRangeForms() {
+        listOf(
+            "0000-01-01T00:00:00Z", "2026-02-29T00:00:00Z",
+            "2026-09-05 18:34:12+02:00", "20260905T183412+0200",
+            "2026-W36-5T18:34:12+02:00", "2026-09-05T18:34:12,5+02:00",
+            "2026-09-05T18:34:12+0200", "2026-09-05T18:34:12+02",
+            "2026-09-05T18:34.5Z", "2026-09-05T18:34:60Z",
+            "2026-09-05T18:34:12+24:00",
+        ).forEach { assertNull(it, TrainlogTimestamp.parse(it)) }
+    }
+}

catalog/joint-actions-v1.json

@@ -0,0 +1,808 @@
+{
+  "format": "trainlog-joint-actions-v1",
+  "version": 1,
+  "joint_actions": [
+    {
+      "action_id": "ankle_dorsiflexion",
+      "definition": "Bring dorsum of foot toward shin",
+      "anatomical_region": "ankle",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Dorsiflexion de la cheville",
+      "joint_complex": "talocrural_joint",
+      "joint_or_complex": "talocrural_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "tibialis_anterior"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "ankle_plantarflexion",
+      "definition": "Point foot away from shin",
+      "anatomical_region": "ankle",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion plantaire de la cheville",
+      "joint_complex": "talocrural_joint",
+      "joint_or_complex": "talocrural_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "gastrocnemius",
+        "soleus"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "elbow_extension",
+      "definition": "Increase elbow angle",
+      "anatomical_region": "elbow",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension du coude",
+      "joint_complex": "humeroulnar_humeroradial_complex",
+      "joint_or_complex": "humeroulnar_humeroradial_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "triceps_brachii"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "elbow_flexion",
+      "definition": "Reduce elbow angle",
+      "anatomical_region": "elbow",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion du coude",
+      "joint_complex": "humeroulnar_humeroradial_complex",
+      "joint_or_complex": "humeroulnar_humeroradial_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "biceps_brachii",
+        "brachialis",
+        "brachioradialis"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "finger_flexion",
+      "definition": "Close fingers toward palm",
+      "anatomical_region": "finger",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion des doigts",
+      "joint_complex": "metacarpophalangeal_and_interphalangeal_joints",
+      "joint_or_complex": "metacarpophalangeal_and_interphalangeal_joints",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "forearm_flexors"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "forearm_pronation",
+      "definition": "Rotate forearm toward palm-back orientation",
+      "anatomical_region": "forearm",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Pronation de l’avant-bras",
+      "joint_complex": "proximal_and_distal_radioulnar_joints",
+      "joint_or_complex": "proximal_and_distal_radioulnar_joints",
+      "principal_plane": "transverse",
+      "plane_notes": "Axial rotation of forearm; spatial plane depends on forearm position.",
+      "contributing_muscle_ids": [
+        "forearm_pronators"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "forearm_supination",
+      "definition": "Rotate forearm toward palm-forward orientation",
+      "anatomical_region": "forearm",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Supination de l’avant-bras",
+      "joint_complex": "proximal_and_distal_radioulnar_joints",
+      "joint_or_complex": "proximal_and_distal_radioulnar_joints",
+      "principal_plane": "transverse",
+      "plane_notes": "Axial rotation of forearm; spatial plane depends on forearm position.",
+      "contributing_muscle_ids": [
+        "biceps_brachii",
+        "supinator"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_abduction",
+      "definition": "Move femur away from midline",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_gluteus_minimus",
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Abduction de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "frontal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "gluteus_medius",
+        "gluteus_minimus",
+        "tensor_fasciae_latae"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_adduction",
+      "definition": "Move femur toward midline",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Adduction de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "frontal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "adductor_magnus",
+        "hip_adductors"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_extension",
+      "definition": "Move femur posteriorly or rise from flexion",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "maeo_2021",
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "adductor_magnus",
+        "gluteus_maximus",
+        "hamstrings_biarticular"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_external_rotation",
+      "definition": "Rotate femur outward",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Rotation externe de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "gluteus_maximus"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_flexion",
+      "definition": "Bring femur toward anterior trunk",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_gluteus_minimus",
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "iliopsoas",
+        "rectus_femoris",
+        "tensor_fasciae_latae"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "hip_internal_rotation",
+      "definition": "Rotate femur inward",
+      "anatomical_region": "hip",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_gluteus_minimus",
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Rotation interne de la hanche",
+      "joint_complex": "acetabulofemoral_joint",
+      "joint_or_complex": "acetabulofemoral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "gluteus_minimus",
+        "tensor_fasciae_latae"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "knee_extension",
+      "definition": "Increase knee angle",
+      "anatomical_region": "knee",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension du genou",
+      "joint_complex": "tibiofemoral_complex",
+      "joint_or_complex": "tibiofemoral_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "quadriceps",
+        "rectus_femoris",
+        "vasti"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "knee_flexion",
+      "definition": "Reduce knee angle",
+      "anatomical_region": "knee",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "maeo_2021",
+        "openstax_actions",
+        "openstax_lower"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion du genou",
+      "joint_complex": "tibiofemoral_complex",
+      "joint_or_complex": "tibiofemoral_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "biceps_femoris_short_head",
+        "gastrocnemius",
+        "hamstrings",
+        "hamstrings_biarticular"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_depression",
+      "definition": "Move scapula inferiorly",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Abaissement scapulaire",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "pectoralis_minor",
+        "trapezius_lower"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_downward_rotation",
+      "definition": "Rotate glenoid downward on return",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Rotation scapulaire vers le bas",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "rhomboids"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_elevation",
+      "definition": "Move scapula superiorly",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Élévation scapulaire",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "trapezius_upper"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_protraction",
+      "definition": "Move scapula anterolaterally around thorax",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Protraction scapulaire",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "pectoralis_minor",
+        "serratus_anterior"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_retraction",
+      "definition": "Move scapula toward vertebral column",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Rétraction scapulaire",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "rhomboids",
+        "trapezius_middle"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "scapular_upward_rotation",
+      "definition": "Rotate glenoid upward during arm elevation",
+      "anatomical_region": "scapular",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Scapular actions are scapulothoracic descriptions produced through coordinated shoulder-girdle articulations.",
+      "display_name_fr": "Rotation scapulaire vers le haut",
+      "joint_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "joint_or_complex": "scapulothoracic_function_via_sternoclavicular_and_acromioclavicular_joints",
+      "principal_plane": "three_dimensional_scapular_motion",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "serratus_anterior",
+        "trapezius_lower",
+        "trapezius_upper"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_abduction",
+      "definition": "Move humerus away from trunk in frontal/scapular plane",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Abduction de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "frontal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_middle",
+        "supraspinatus"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_adduction",
+      "definition": "Move humerus toward trunk",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_pectoralis",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Adduction de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "frontal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "latissimus_dorsi",
+        "pectoralis_major",
+        "pectoralis_major_sternocostal",
+        "teres_major"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_extension",
+      "definition": "Move humerus posteriorly or return from flexion",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "nih_pectoralis",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_posterior",
+        "latissimus_dorsi",
+        "pectoralis_major_sternocostal",
+        "teres_major"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_external_rotation",
+      "definition": "Rotate humerus outward about long axis",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Rotation externe de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_posterior",
+        "infraspinatus",
+        "teres_minor"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_flexion",
+      "definition": "Move humerus anteriorly/elevate forward relative to trunk",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "nih_pectoralis",
+        "openstax_actions"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_anterior",
+        "pectoralis_major_clavicular"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_horizontal_abduction",
+      "definition": "Move elevated humerus backward in transverse plane",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "openstax_actions"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Abduction horizontale de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_posterior"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_horizontal_adduction",
+      "definition": "Bring elevated humerus across anterior trunk",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "nih_pectoralis",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Adduction horizontale de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_anterior",
+        "pectoralis_major",
+        "pectoralis_major_clavicular",
+        "pectoralis_major_sternocostal"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "shoulder_internal_rotation",
+      "definition": "Rotate humerus inward about long axis",
+      "anatomical_region": "shoulder",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_deltoid",
+        "nih_pectoralis",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Rotation interne de l’épaule",
+      "joint_complex": "glenohumeral_joint",
+      "joint_or_complex": "glenohumeral_joint",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "deltoid_anterior",
+        "latissimus_dorsi",
+        "pectoralis_major",
+        "subscapularis",
+        "teres_major"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "trunk_extension",
+      "definition": "Extend vertebral column; not synonymous with hip extension",
+      "anatomical_region": "trunk",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_back"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension du tronc",
+      "joint_complex": "intervertebral_column_complex",
+      "joint_or_complex": "intervertebral_column_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "erector_spinae",
+        "multifidus"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "trunk_flexion",
+      "definition": "Flex vertebral column; not synonymous with hip flexion",
+      "anatomical_region": "trunk",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_trunk"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion du tronc",
+      "joint_complex": "intervertebral_column_complex",
+      "joint_or_complex": "intervertebral_column_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "external_oblique",
+        "internal_oblique",
+        "rectus_abdominis"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "trunk_lateral_flexion",
+      "definition": "Bend vertebral column sideways",
+      "anatomical_region": "trunk",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_back",
+        "openstax_trunk"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Inclinaison latérale du tronc",
+      "joint_complex": "intervertebral_column_complex",
+      "joint_or_complex": "intervertebral_column_complex",
+      "principal_plane": "frontal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "erector_spinae",
+        "external_oblique",
+        "internal_oblique",
+        "quadratus_lumborum"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "trunk_rotation",
+      "definition": "Rotate thorax and pelvis relative to each other",
+      "anatomical_region": "trunk",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_back",
+        "openstax_trunk"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Rotation du tronc",
+      "joint_complex": "intervertebral_column_complex",
+      "joint_or_complex": "intervertebral_column_complex",
+      "principal_plane": "transverse",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "external_oblique",
+        "internal_oblique",
+        "multifidus"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "wrist_extension",
+      "definition": "Bend dorsum of hand toward posterior forearm",
+      "anatomical_region": "wrist",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Extension du poignet",
+      "joint_complex": "radiocarpal_midcarpal_complex",
+      "joint_or_complex": "radiocarpal_midcarpal_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "forearm_extensors"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    },
+    {
+      "action_id": "wrist_flexion",
+      "definition": "Bend palm toward anterior forearm",
+      "anatomical_region": "wrist",
+      "evidence_type": "established_anatomy",
+      "confidence": "high",
+      "source_refs": [
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Opposite motion in a loaded return may be controlled eccentrically by the same agonists.",
+      "display_name_fr": "Flexion du poignet",
+      "joint_complex": "radiocarpal_midcarpal_complex",
+      "joint_or_complex": "radiocarpal_midcarpal_complex",
+      "principal_plane": "sagittal",
+      "plane_notes": "Nominal anatomical plane; actual exercise can combine planes. Scapular kinematics are three-dimensional.",
+      "contributing_muscle_ids": [
+        "forearm_flexors"
+      ],
+      "contributor_semantics": "Non-exhaustive anatomical capability list; not a primary-role list or inverse EMG ranking. Aggregate/member overlap retained explicitly."
+    }
+  ]
+}

catalog/movement-patterns-v1.json

@@ -0,0 +1,538 @@
+{
+  "format": "trainlog-movement-patterns-v1",
+  "version": 1,
+  "movement_patterns": [
+    {
+      "pattern_id": "cyclic_lower_limb",
+      "definition": "Repeated lower-limb propulsion such as cycling; not identical to gait.",
+      "typical_action_ids": [
+        "hip_extension",
+        "hip_flexion",
+        "knee_extension",
+        "knee_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Propulsion cyclique des membres inférieurs",
+      "typical_body_zone_ids": [
+        "calves",
+        "glutes",
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "cyclic_rowing",
+      "definition": "Repeated leg drive, trunk control and arm pull; cardio rower differs from seated resistance row.",
+      "typical_action_ids": [
+        "elbow_flexion",
+        "hip_extension",
+        "knee_extension",
+        "shoulder_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Rame cyclique",
+      "typical_body_zone_ids": [
+        "arms",
+        "back",
+        "core",
+        "glutes",
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "hip_dominant",
+      "definition": "Multi-joint task organized around hip extension with comparatively restrained knee excursion.",
+      "typical_action_ids": [
+        "hip_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Mouvement à dominante hanche",
+      "typical_body_zone_ids": [
+        "glutes",
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "horizontal_pull",
+      "definition": "Pull toward torso with shoulder extension/horizontal abduction and elbow flexion.",
+      "typical_action_ids": [
+        "elbow_flexion",
+        "scapular_retraction",
+        "shoulder_extension",
+        "shoulder_horizontal_abduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Tirage horizontal",
+      "typical_body_zone_ids": [
+        "arms",
+        "back"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "horizontal_push",
+      "definition": "Press resistance away anterior to torso with coordinated shoulder and elbow movement.",
+      "typical_action_ids": [
+        "elbow_extension",
+        "shoulder_horizontal_adduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Poussée horizontale",
+      "typical_body_zone_ids": [
+        "arms",
+        "chest",
+        "shoulders"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "knee_dominant",
+      "definition": "Multi-joint lower-limb task with substantial knee-extension demand; hip extension remains involved.",
+      "typical_action_ids": [
+        "hip_extension",
+        "knee_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Mouvement à dominante genou",
+      "typical_body_zone_ids": [
+        "glutes",
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "locomotion",
+      "definition": "Repeated gait cycles with support and progression; treadmill belt can replace overground translation.",
+      "typical_action_ids": [
+        "ankle_dorsiflexion",
+        "ankle_plantarflexion",
+        "hip_extension",
+        "hip_flexion",
+        "knee_extension",
+        "knee_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Locomotion",
+      "typical_body_zone_ids": [
+        "calves",
+        "glutes",
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "shoulder_abduction",
+      "display_name_fr": "Abduction de l’épaule",
+      "definition": "Raise humerus laterally or in scapular plane against resistance; lateral-raise family. Scapular upward rotation accompanies larger elevation.",
+      "typical_action_ids": [
+        "scapular_upward_rotation",
+        "shoulder_abduction"
+      ],
+      "typical_body_zone_ids": [
+        "shoulders"
+      ],
+      "parent_pattern_id": null,
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "nih_deltoid",
+        "openstax_actions",
+        "openstax_upper"
+      ],
+      "notes": "Explicit authored task convention; anti-motion is a demand, not a new joint action or claim of muscle isolation."
+    },
+    {
+      "pattern_id": "single_joint_ankle_plantarflexion",
+      "definition": "Predominant resisted ankle plantarflexion.",
+      "typical_action_ids": [
+        "ankle_plantarflexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Flexion plantaire",
+      "typical_body_zone_ids": [
+        "calves"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_elbow_extension",
+      "definition": "Predominant resisted elbow extension.",
+      "typical_action_ids": [
+        "elbow_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Extension isolée du coude",
+      "typical_body_zone_ids": [
+        "arms"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_elbow_flexion",
+      "definition": "Predominant resisted elbow flexion.",
+      "typical_action_ids": [
+        "elbow_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Flexion isolée du coude",
+      "typical_body_zone_ids": [
+        "arms"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_hip_abduction",
+      "definition": "Predominant resisted hip abduction.",
+      "typical_action_ids": [
+        "hip_abduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Abduction isolée de hanche",
+      "typical_body_zone_ids": [
+        "glutes"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_hip_adduction",
+      "definition": "Predominant resisted hip adduction.",
+      "typical_action_ids": [
+        "hip_adduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Adduction isolée de hanche",
+      "typical_body_zone_ids": [
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_knee_extension",
+      "definition": "Predominant resisted knee extension.",
+      "typical_action_ids": [
+        "knee_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Extension isolée du genou",
+      "typical_body_zone_ids": [
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_knee_flexion",
+      "definition": "Predominant resisted knee flexion.",
+      "typical_action_ids": [
+        "knee_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Flexion isolée du genou",
+      "typical_body_zone_ids": [
+        "thighs"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_shoulder_horizontal_abduction",
+      "definition": "Reverse-fly humeral horizontal abduction with approximately fixed elbow.",
+      "typical_action_ids": [
+        "shoulder_horizontal_abduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Écarté inversé",
+      "typical_body_zone_ids": [
+        "back",
+        "shoulders"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "single_joint_shoulder_horizontal_adduction",
+      "definition": "Fly-like humeral adduction across torso with approximately fixed elbow.",
+      "typical_action_ids": [
+        "shoulder_horizontal_adduction"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Écarté pectoral",
+      "typical_body_zone_ids": [
+        "chest",
+        "shoulders"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "trunk_anti_extension",
+      "display_name_fr": "Résistance à l’extension du tronc",
+      "definition": "Resist an external trunk-extension moment while maintaining intended spinal orientation. No dynamic joint action is required.",
+      "typical_action_ids": [],
+      "typical_body_zone_ids": [
+        "core"
+      ],
+      "parent_pattern_id": "trunk_stabilization",
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_back"
+      ],
+      "notes": "Explicit authored task convention; anti-motion is a demand, not a new joint action or claim of muscle isolation."
+    },
+    {
+      "pattern_id": "trunk_anti_rotation",
+      "display_name_fr": "Résistance à la rotation du tronc",
+      "definition": "Resist an external rotational moment between thorax and pelvis while maintaining intended orientation.",
+      "typical_action_ids": [],
+      "typical_body_zone_ids": [
+        "core"
+      ],
+      "parent_pattern_id": "trunk_stabilization",
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_back"
+      ],
+      "notes": "Explicit authored task convention; anti-motion is a demand, not a new joint action or claim of muscle isolation."
+    },
+    {
+      "pattern_id": "trunk_extension",
+      "definition": "Dynamic resisted spinal extension.",
+      "typical_action_ids": [
+        "trunk_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Extension du tronc",
+      "typical_body_zone_ids": [
+        "back",
+        "core"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "trunk_flexion",
+      "definition": "Dynamic resisted spinal flexion.",
+      "typical_action_ids": [
+        "trunk_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Flexion du tronc",
+      "typical_body_zone_ids": [
+        "core"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "trunk_lateral_stability",
+      "display_name_fr": "Stabilité latérale du tronc",
+      "definition": "Resist an external lateral-flexion moment while maintaining trunk orientation; not dynamic side bending.",
+      "typical_action_ids": [],
+      "typical_body_zone_ids": [
+        "back",
+        "core"
+      ],
+      "parent_pattern_id": "trunk_stabilization",
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "nih_abdominal_wall",
+        "openstax_actions",
+        "openstax_back"
+      ],
+      "notes": "Explicit authored task convention; anti-motion is a demand, not a new joint action or claim of muscle isolation."
+    },
+    {
+      "pattern_id": "trunk_rotation",
+      "definition": "Dynamic resisted thorax-pelvis rotation.",
+      "typical_action_ids": [
+        "trunk_rotation"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Rotation du tronc",
+      "typical_body_zone_ids": [
+        "core"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "trunk_stabilization",
+      "definition": "Maintain trunk orientation against external moments; anti-extension, anti-rotation and anti-lateral-flexion are demands, not anatomical joint motions.",
+      "typical_action_ids": [],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Stabilisation du tronc",
+      "typical_body_zone_ids": [
+        "back",
+        "core"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "vertical_pull",
+      "definition": "Pull from overhead toward torso.",
+      "typical_action_ids": [
+        "elbow_flexion",
+        "shoulder_adduction",
+        "shoulder_extension"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Tirage vertical",
+      "typical_body_zone_ids": [
+        "arms",
+        "back"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    },
+    {
+      "pattern_id": "vertical_push",
+      "definition": "Press overhead relative to torso with humeral elevation and elbow extension.",
+      "typical_action_ids": [
+        "elbow_extension",
+        "scapular_upward_rotation",
+        "shoulder_abduction",
+        "shoulder_flexion"
+      ],
+      "evidence_type": "practical_inference",
+      "confidence": "moderate",
+      "source_refs": [
+        "acsm_2009",
+        "openstax_actions"
+      ],
+      "notes": "Authored classification convention; typical actions are not mandatory in every variant and single-joint does not mean one muscle or zero stabilization.",
+      "display_name_fr": "Poussée verticale",
+      "typical_body_zone_ids": [
+        "arms",
+        "shoulders"
+      ],
+      "body_zone_semantics": "Descriptive common participation, not a persisted exercise mapping or requirement that every listed zone be assigned."
+    }
+  ]
+}

catalog/science-references-v1.json

@@ -0,0 +1,407 @@
+{
+  "format": "trainlog-science-references-v1",
+  "version": 1,
+  "references": [
+    {
+      "ref_id": "acsm_2009",
+      "title": "American College of Sports Medicine position stand. Progression models in resistance training for healthy adults.",
+      "year": 2009,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/19204579/",
+      "doi": "10.1249/mss.0b013e3181915670",
+      "pmid": "19204579",
+      "topics": [
+        "progression",
+        "specificity",
+        "programming"
+      ],
+      "notes": "Progressive overload and specificity organize resistance-training progression; adjustment should follow achieved performance and goals.",
+      "limitations": "Historical professional position stand; contemporary synthesis above takes precedence for comparative outcomes. No numeric recommendations adopted.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "American College of Sports Medicine.",
+      "type": "practical_inference"
+    },
+    {
+      "ref_id": "currier_2023",
+      "title": "Resistance training prescription for muscle strength and hypertrophy in healthy adults: a systematic review and Bayesian network meta-analysis.",
+      "year": 2023,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/37414459/",
+      "doi": "10.1136/bjsports-2023-106807",
+      "pmid": "37414459",
+      "topics": [
+        "programming",
+        "load",
+        "sets",
+        "frequency"
+      ],
+      "notes": "Across reviewed adult trials, resistance training improved strength and hypertrophy; heavier loads ranked better for strength, and multiple sets characterized higher-ranked hypertrophy programs.",
+      "limitations": "Network rankings are not individual prescriptions or proof of one universally optimal combination.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Currier BS, Mcleod JC, Banfield L, Beyene J, Welton NJ, D'Souza AC, Keogh JAJ, Lin L, Coletta G, Yang A, Colenso-Semple L, Lau KJ, Verboom A, Phillips SM.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "franke_2015",
+      "title": "Analysis of anterior, middle and posterior deltoid activation during single and multijoint exercises.",
+      "year": 2015,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/24947920/",
+      "doi": null,
+      "pmid": "24947920",
+      "topics": [
+        "rear_delt",
+        "rows"
+      ],
+      "notes": "Reverse pec-deck elicited greater posterior-deltoid surface EMG than the studied row/pulldown tasks.",
+      "limitations": "Twelve trained men; selected exercises/techniques only; no longitudinal hypertrophy comparison; no DOI reported in indexed record.",
+      "accessed_on": "2026-09-09",
+      "publication_note": "Print 2015; online indexed June 2014.",
+      "authors_or_organization": "Franke Rde A, Botton CE, Rodrigues R, Pinto RS, Lima CS.",
+      "type": "emg_evidence"
+    },
+    {
+      "ref_id": "grgic_2022",
+      "title": "Effects of resistance training performed to repetition failure or non-failure on muscular strength and hypertrophy: A systematic review and meta-analysis.",
+      "year": 2022,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/33497853/",
+      "doi": "10.1016/j.jshs.2021.01.007",
+      "pmid": "33497853",
+      "topics": [
+        "failure",
+        "strength",
+        "hypertrophy"
+      ],
+      "notes": "Failure was not generally necessary for strength or hypertrophy gains in included comparisons.",
+      "limitations": "Nonfailure effort and volume differ between studies; this does not imply arbitrary easy sets equal hard sets.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Grgic J, Schoenfeld BJ, Orazem J, Sabol F.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "kassiano_2022",
+      "title": "Does Varying Resistance Exercises Promote Superior Muscle Hypertrophy and Strength Gains? A Systematic Review.",
+      "year": 2022,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/35438660/",
+      "doi": "10.1519/jsc.0000000000004258",
+      "pmid": "35438660",
+      "topics": [
+        "selection",
+        "variation"
+      ],
+      "notes": "Systematic exercise variation may distribute regional stimulus and support task-specific strength; excessive random variation lacks support.",
+      "limitations": "Eight studies, all young men; limited generalizability and no universal rotation schedule.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Kassiano W, Nunes JP, Costa B, Ribeiro AS, Schoenfeld BJ, Cyrino ES.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "lee_2015",
+      "title": "Enhanced muscle activity during lumbar extension exercise with pelvic stabilization.",
+      "year": 2015,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/26730390/",
+      "doi": "10.12965/jer.150249",
+      "pmid": "26730390",
+      "topics": [
+        "back_extension",
+        "pelvic_stabilization"
+      ],
+      "notes": "Pelvic stabilization altered lumbar-extensor excitation during extension testing.",
+      "limitations": "Acute small study; equipment restraint and execution matter; no claim of clinical benefit or universal isolation.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Lee HS.",
+      "type": "emg_evidence"
+    },
+    {
+      "ref_id": "lehman_2004",
+      "title": "Variations in muscle activation levels during traditional latissimus dorsi weight training exercises: An experimental study.",
+      "year": 2004,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/15228624/",
+      "doi": "10.1186/1476-5918-3-4",
+      "pmid": "15228624",
+      "topics": [
+        "lat_pulldown",
+        "seated_row"
+      ],
+      "notes": "Measured latissimus, elbow-flexor and scapular-muscle activity differed across studied pulldown/row tasks.",
+      "limitations": "Isometric portions and surface recordings; no mapping of all divergent machine trajectories or outcome superiority.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Lehman GJ, Buchan DD, Lundy A, Myers N, Nalborczyk A.",
+      "type": "emg_evidence"
+    },
+    {
+      "ref_id": "life_fitness_catalog_2024",
+      "title": "Life Fitness commercial product catalogue 2024",
+      "year": 2024,
+      "url": "https://www.lifefitness.com.au/wp-content/uploads/2024/02/Life-Fitness-Catalogue_2024_web.pdf",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "multifunction",
+        "equipment_identity"
+      ],
+      "notes": "Catalogue lists distinct Pectoral Fly/Rear Deltoid and Assist Dip Chin combination products.",
+      "limitations": "EXAMPLE ONLY: establishes available equipment concepts, not manufacturer/model identification in the observed gym or anatomical efficacy.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Life Fitness",
+      "type": "manufacturer_statement"
+    },
+    {
+      "ref_id": "maeo_2021",
+      "title": "Greater Hamstrings Muscle Hypertrophy but Similar Damage Protection after Training at Long versus Short Muscle Lengths.",
+      "year": 2021,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/33009197/",
+      "doi": "10.1249/mss.0000000000002523",
+      "pmid": "33009197",
+      "topics": [
+        "hamstrings",
+        "seated_leg_curl",
+        "prone_leg_curl"
+      ],
+      "notes": "Within-person training found greater biarticular hamstring growth with seated than prone curls; both variants trained knee flexion.",
+      "limitations": "One protocol/population; not a guarantee for every machine, person, muscle region or strength task.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Maeo S, Huang M, Wu Y, Sakurai H, Kusagawa Y, Sugiyama T, Kanehisa H, Isaka T.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "martin_fuentes_2020",
+      "title": "Evaluation of the Lower Limb Muscles' Electromyographic Activity during the Leg Press Exercise and Its Variants: A Systematic Review.",
+      "year": 2020,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/32605065/",
+      "doi": "10.3390/ijerph17134626",
+      "pmid": "32605065",
+      "topics": [
+        "leg_press"
+      ],
+      "notes": "Leg-press studies report substantial quadriceps excitation; variant findings are inconsistent.",
+      "limitations": "EMG review does not establish universal foot-position targeting, force shares or hypertrophy ranking.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Martín-Fuentes I, Oliva-Lozano JM, Muyor JM.",
+      "type": "emg_evidence"
+    },
+    {
+      "ref_id": "nih_abdominal_wall",
+      "title": "Anatomy, Abdomen and Pelvis: Anterolateral Abdominal Wall",
+      "year": 2023,
+      "url": "https://www.ncbi.nlm.nih.gov/books/NBK525975/",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "core",
+        "trunk"
+      ],
+      "notes": "Rectus abdominis, internal/external obliques and transversus have movement, abdominal-wall tension and stabilization roles.",
+      "limitations": "Functional anatomy only; clinical sections are outside this handoff scope.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Kevin Seeras; Ryan N. Qasawa; Ricky Ju; Shivana Prakash",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "nih_deltoid",
+      "title": "Anatomy, Shoulder and Upper Limb, Deltoid Muscle",
+      "year": 2024,
+      "url": "https://www.ncbi.nlm.nih.gov/books/NBK537056/",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "deltoids",
+        "shoulder"
+      ],
+      "notes": "Anterior, middle and posterior portions have distinct lines of action; relative contribution depends on arm position.",
+      "limitations": "Functional anatomy only; clinical sections are outside this handoff scope.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Adel Elzanie; Matthew A. Varacallo",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "nih_gluteus_minimus",
+      "title": "Anatomy, Bony Pelvis and Lower Limb, Gluteus Minimus Muscle",
+      "year": 2023,
+      "url": "https://www.ncbi.nlm.nih.gov/books/NBK556144/",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "hip_abduction",
+        "glutes",
+        "gait"
+      ],
+      "notes": "Minimus abducts and stabilizes the hip; anterior fibers support internal rotation. Prevents erroneous adduction classification from textbook alternative text.",
+      "limitations": "Functional anatomy only; clinical sections are outside this handoff scope.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Anthony J. Greco; Renato C. Vilella",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "nih_pectoralis",
+      "title": "Anatomy, Thorax, Pectoralis Major Major",
+      "year": 2023,
+      "url": "https://www.ncbi.nlm.nih.gov/books/NBK525991/",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "chest",
+        "shoulder"
+      ],
+      "notes": "Clavicular and sternocostal parts share shoulder adduction/internal rotation; clavicular fibers support flexion, sternocostal fibers extension from flexion. Regional anatomy does not prove isolated regional training.",
+      "limitations": "Functional anatomy only; clinical sections are outside this handoff scope.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Francesca Solari; Bracken Burns",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "openstax_actions",
+      "title": "Anatomy and Physiology 2e: Types of Body Movements",
+      "year": 2022,
+      "url": "https://openstax.org/books/anatomy-and-physiology-2e/pages/9-5-types-of-body-movements",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "joint_actions"
+      ],
+      "notes": "Terminology of joint movements. Pattern labels in Trainlog remain authored programming abstractions.",
+      "limitations": "Textbook synthesis; not a machine-specific force or adaptation study.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "J. Gordon Betts; Kelly A. Young; James A. Wise; Eddie Johnson; Brandon Poe; Dean H. Kruse; Oksana Korol; Jody E. Johnson; Mark Womble; Peter DeSaix",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "openstax_back",
+      "title": "Anatomy and Physiology 2e: Axial Muscles of the Head, Neck, and Back",
+      "year": 2022,
+      "url": "https://openstax.org/books/anatomy-and-physiology-2e/pages/11-3-axial-muscles-of-the-head-neck-and-back",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "spinal_extensors",
+        "multifidus"
+      ],
+      "notes": "Vertebral-column muscle functions including erector spinae and deep posterior muscles.",
+      "limitations": "Textbook synthesis; not a machine-specific force or adaptation study.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "J. Gordon Betts; Kelly A. Young; James A. Wise; Eddie Johnson; Brandon Poe; Dean H. Kruse; Oksana Korol; Jody E. Johnson; Mark Womble; Peter DeSaix",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "openstax_lower",
+      "title": "Anatomy and Physiology 2e: Appendicular Muscles of the Pelvic Girdle and Lower Limbs",
+      "year": 2022,
+      "url": "https://openstax.org/books/anatomy-and-physiology-2e/pages/11-6-appendicular-muscles-of-the-pelvic-girdle-and-lower-limbs",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "hip",
+        "knee",
+        "ankle"
+      ],
+      "notes": "Functional lower-limb anatomy. Figure alternative text contains direction inconsistencies for minimus/pectineus/gracilis; use action definitions and NIH corroboration, not those phrases.",
+      "limitations": "Textbook synthesis; not a machine-specific force or adaptation study.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "J. Gordon Betts; Kelly A. Young; James A. Wise; Eddie Johnson; Brandon Poe; Dean H. Kruse; Oksana Korol; Jody E. Johnson; Mark Womble; Peter DeSaix",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "openstax_trunk",
+      "title": "Anatomy and Physiology 2e: Axial Muscles of the Abdominal Wall and Thorax",
+      "year": 2022,
+      "url": "https://openstax.org/books/anatomy-and-physiology-2e/pages/11-4-axial-muscles-of-the-abdominal-wall-and-thorax",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "trunk",
+        "abdominals",
+        "spinal_extensors"
+      ],
+      "notes": "Trunk movement, abdominal compression and postural roles.",
+      "limitations": "Textbook synthesis; not a machine-specific force or adaptation study.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "J. Gordon Betts; Kelly A. Young; James A. Wise; Eddie Johnson; Brandon Poe; Dean H. Kruse; Oksana Korol; Jody E. Johnson; Mark Womble; Peter DeSaix",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "openstax_upper",
+      "title": "Anatomy and Physiology 2e: Muscles of the Pectoral Girdle and Upper Limbs",
+      "year": 2022,
+      "url": "https://openstax.org/books/anatomy-and-physiology-2e/pages/11-5-muscles-of-the-pectoral-girdle-and-upper-limbs",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "shoulder",
+        "scapula",
+        "elbow",
+        "forearm"
+      ],
+      "notes": "Functional anatomy of upper-limb and scapular muscles; no exercise outcome ranking.",
+      "limitations": "Textbook synthesis; not a machine-specific force or adaptation study.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "J. Gordon Betts; Kelly A. Young; James A. Wise; Eddie Johnson; Brandon Poe; Dean H. Kruse; Oksana Korol; Jody E. Johnson; Mark Womble; Peter DeSaix",
+      "type": "established_anatomy"
+    },
+    {
+      "ref_id": "precor_pulley_example",
+      "title": "Resolute Dual Adjustable Pulley RUD0915",
+      "year": null,
+      "url": "https://www.precor.com/en-US/products/RUD0915",
+      "doi": null,
+      "pmid": null,
+      "topics": [
+        "cable",
+        "load_context"
+      ],
+      "notes": "Manufacturer documents a 4:1 cable ratio on this particular example, demonstrating why stack labels and handle resistance differ.",
+      "limitations": "EXAMPLE ONLY: not evidence that the user owns Precor or this model; local pulley ratios remain unresolved.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Precor",
+      "type": "manufacturer_statement"
+    },
+    {
+      "ref_id": "refalo_2023",
+      "title": "Influence of Resistance Training Proximity-to-Failure on Skeletal Muscle Hypertrophy: A Systematic Review with Meta-analysis.",
+      "year": 2023,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/36334240/",
+      "doi": "10.1007/s40279-022-01784-y",
+      "pmid": "36334240",
+      "topics": [
+        "proximity_to_failure"
+      ],
+      "notes": "Failure definitions matter; available categorical comparisons do not establish a simple more-failure-is-better hypertrophy rule.",
+      "limitations": "Literature search predates newer trials; uncertainty in actual repetitions in reserve prevents exact universal thresholds.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Refalo MC, Helms ER, Trexler ET, Hamilton DL, Fyfe JJ.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "vieira_2022",
+      "title": "Effects of Resistance Training to Muscle Failure on Acute Fatigue: A Systematic Review and Meta-Analysis.",
+      "year": 2022,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/34881412/",
+      "doi": "10.1007/s40279-021-01602-x",
+      "pmid": "34881412",
+      "topics": [
+        "fatigue",
+        "recovery",
+        "failure"
+      ],
+      "notes": "Failure training generated greater acute fatigue than nonfailure conditions in the reviewed experiments.",
+      "limitations": "Acute markers do not establish a universal recovery duration or predict individual long-term adaptation.",
+      "accessed_on": "2026-09-09",
+      "authors_or_organization": "Vieira JG, Sardeli AV, Dias MR, Filho JE, Campos Y, Sant'Ana L, Leitão L, Reis V, Wilk M, Novaes J, Vianna J.",
+      "type": "intervention_evidence"
+    },
+    {
+      "ref_id": "vigotsky_2018",
+      "title": "Interpreting Signal Amplitudes in Surface Electromyography Studies in Sport and Rehabilitation Sciences.",
+      "year": 2017,
+      "url": "https://pubmed.ncbi.nlm.nih.gov/29354060/",
+      "doi": "10.3389/fphys.2017.00985",
+      "pmid": "29354060",
+      "topics": [
+        "measurement",
+        "emg"
+      ],
+      "notes": "Surface EMG amplitude reflects a recording-dependent excitation signal and cannot directly rank hypertrophy, force or training effectiveness.",
+      "limitations": "Methodological review; anatomical inference and longitudinal outcomes require separate evidence.",
+      "accessed_on": "2026-09-09",
+      "publication_note": "Volume/index year 2017; online publication January 4, 2018.",
+      "authors_or_organization": "Vigotsky AD, Halperin I, Lehman GJ, Trajano GS, Vieira TM.",
+      "type": "emg_evidence"
+    }
+  ]
+}

docs/android.md

@@ -510,3 +510,28 @@ Android requires `data_fields = 0` for `SETS`, while the desktop model/API
 currently accepts known supplemental bits on either recording mode. Supplied
 profiles do not exercise this difference. Supporting a future set-based
 supplemental field requires an explicit shared-model decision.
+
+## 18. Training knowledge V1 read APIs
+
+Android bundles the six authored training-knowledge catalogs as immutable
+assets. `TrainingKnowledgeCatalog` validates and exposes their stable-ID
+records; it is not a second manually authored scientific table.
+`TrainlogRepository.getTrainingExerciseContext()` composes an exact persisted
+exercise with its direct/ancestor persisted zones, optional scientific mapping,
+compatible equipment, latest explicit MAX and recent occurrence/set preview.
+`listExerciseOccurrences()` and `listExerciseOccurrenceSets()` provide bounded
+follow-up pages. Occurrence and set limits are 1–32 and 1–64 respectively.
+Cursors order current data chronologically by original timestamp, session ID
+and occurrence ID, and do not preserve a snapshot across calls.
+
+This read-only feature makes no Android schema change (the runtime schema
+remains v10), does not seed rows, and does not export/synchronize new data. It
+does not implement recommendations, planned weights, set counts or fatigue
+scores. Its occurrence and latest-MAX readers use the same explicit temporal
+grammar, exact fractional comparison and bytewise ID tie breakers as C.
+The Android writer's omitted-seconds form is admitted, and emitted cursors
+retain the original source text. Production pagination/MAX parity tests pass.
+The tranche is `TRAINING_KNOWLEDGE_V1=PASS`. The Android loader enforces canonical exercise
+and equipment identity syntax, bidirectional exercise/capability compatibility,
+HIGH evidence source type, and non-unresolved BODY ZONE audit evidence. Its
+full source and uncertainty contract is in [Training knowledge system V1](domain/knowledge_system.md).

docs/architecture.md

@@ -477,3 +477,36 @@ masks to be zero for `SETS`. The shipped catalog uses supplemental speed and
 distance only with `CONTINUOUS`; defining cross-platform behavior for a future
 set-based supplemental field is a model-contract decision, not part of this
 reconciliation.
+
+## 14. Training knowledge V1 boundary
+
+`TRAINING_KNOWLEDGE_V1` is a read-only composition layer. Six versioned JSON
+catalogs under `catalog/` are the only authored scientific source; generated C
+data and Android asset loading derive from them. They contain cited anatomy,
+movement, exercise and equipment knowledge, not user history. The generated
+knowledge audit is evidence output, not an editable source.
+
+The desktop `training_knowledge.h` API exposes immutable catalog records and
+stable-ID queries. `training_context.h` combines one exact persisted exercise
+with its stored BODY ZONE relations, optional science, compatible equipment,
+latest explicit maximum, and bounded occurrence/set history under one read
+snapshot. Android provides the corresponding catalog and repository context.
+This composition neither writes SQLite nor seeds catalog mappings. An unknown
+runtime ID and a missing scientific record remain valid states.
+
+Scientific BODY ZONE projections and persisted BODY ZONE relations have
+different ownership and are never substituted for one another. Labels and
+generic equipment descriptions are not runtime identities; an equipment
+capability does not create an `ex_<uuid-v4>` exercise or historical association.
+The feature is `TRAINING_KNOWLEDGE_V1=PASS`. The temporal contract has
+independent PASS evidence:
+the readers parse the admitted source forms into exact instants, compare exact
+fractions, then use bytewise session and entry ID ties; emitted exclusive
+cursors preserve the original timestamp text and IDs. Selection and hydration
+share a read snapshot, while separate page calls retain current-data semantics.
+Malformed caller cursors and malformed matching stored timestamps fail
+explicitly; a selected timestamp beyond C's 40-character output field also
+fails explicitly. The initial full-tranche audit's stale temporal-documentation,
+Android loader, Meson input, and C role-only query findings were resolved by one
+bounded repair chain and independently verified. Its full contract, uncertainty boundary and
+future-only planning architecture are in [Training knowledge system V1](domain/knowledge_system.md).

docs/current_state.md

@@ -62,7 +62,9 @@ BODY_ZONES_DESKTOP_REAL_MIGRATION=PASS
 BODY_ZONES_TUI_REAL_VALIDATION=PASS
 BODY_ZONES_ANDROID_DEVICE_VALIDATION=PASS
 
-DESKTOP_TESTS=39/39 PASS
+TRAINING_KNOWLEDGE_V1=PASS
+
+DESKTOP_TESTS=42/42 PASS (recorded validation checkpoint)
 ANDROID_BUILD=PASS
 HARDWARE_SYNC_VALIDATION=HISTORICAL_PASS
 ```
@@ -72,6 +74,8 @@ HARDWARE_SYNC_VALIDATION=HISTORICAL_PASS
 Implemented:
 
 - C17/Notcurses true-color TUI (72x20 minimum, UTF-8 prompts, resize fallback);
+- UTF-8 cell-aware scrolling training-knowledge screen, tested at the 72x20
+  minimum terminal;
 - SQLite schema v11, with stable ordered `session_exercises.entry_id`,
   occurrence-level equipment identity, and desktop-local custom-equipment
   definitions, plus occurrence-owned `max_results`; its v9 -> v10 migration
@@ -152,6 +156,48 @@ All Android screens use the shared compact `◆ TRAINLOG ◆` header: the
 Notcurses accent, muted context line, and flat touch layout reproduce the TUI
 plaque without literal terminal box drawing.
 
+## Training knowledge V1
+
+The implemented read-only training-knowledge layer loads six versioned JSON
+catalogs as the sole authored scientific source, generates the immutable C
+catalog representation, and loads the same assets on Android. It has no
+database migration, no auto-seeding, and no synchronization artifact. The
+desktop database remains schema v11 and Android remains schema v10.
+
+Desktop `training_knowledge.h` and Android `TrainingKnowledgeCatalog` expose
+source-linked science lookups and resolved-candidate filters. Desktop
+`training_context.h` and Android `TrainlogRepository` compose one real runtime
+exercise with its persisted zones, compatible equipment, latest explicit MAX,
+and bounded chronological occurrence/set history. Persisted zones remain
+separate from scientific mappings; missing scientific knowledge is valid.
+
+The scientific review passed and the reviewed catalog bytes remain unchanged.
+The independent temporal review returned `TEMPORAL_DELTA_REVIEW=PASS`, with no
+temporal defects or repairs. It reviewed the grammar, calendar and offset
+bounds, fraction precision, bytewise ties, cursor aliasing and exclusivity,
+source capacity, snapshots, and Android/Python parity; it ran the targeted
+Meson and four Python temporal tests. The initial full-tranche engineering
+audit initially failed with stale temporal-defect documentation (BLOCKER),
+Android loader parity gaps (BLOCKER), omitted Meson generator inputs (BLOCKER),
+and a C role-only query mismatch (HIGH). One bounded repair chain resolved all
+four findings; independent repair verification returned
+`FINAL_REVIEW_REPAIR_VERIFICATION=PASS` and
+`TRAINING_KNOWLEDGE_V1_ENGINEERING_REVIEW=PASS`. Fresh final validation passed:
+strict build, 42 Meson tests, eight knowledge and four temporal Python tests,
+knowledge/JSON/import validators, three C17 headers, affected C knowledge and
+context tests under ASan/UBSan plus Python timestamp validation, normal and
+sanitized 12-form temporal probes, and Android 56 tests with zero failures or
+errors and one known missing-real-v9-fixture skip; Java 17 `assembleDebug` also
+passed. Generated C is byte-identical with SHA-256
+`e8c099f67eb111d61621b5d76592c049823af5508d43e73ec646f22e4c377fca`; all six
+Android assets are byte-identical. Preservation before and after repair confirms
+schema v11/v10, unchanged catalog/science/temporal bytes, and unchanged real
+database logical SHA-256 `26139cafeffbde3ec08f6ef23c5069e75afb9cd69ffffb40be5c099006fedc4d`,
+counts, integrity, and foreign keys. `TRAINING_KNOWLEDGE_V1=PASS`. The
+[temporal contract](reviews/training_knowledge_v1_temporal_contract.md)
+defines the settled reader behavior. No manual Android install, manual TUI
+visual validation or manual MTP validation is claimed for this tranche.
+
 ## Synchronization
 
 Canonical exchange directory:

docs/database.md

@@ -546,3 +546,25 @@ left/right asymmetry percentages
 
 The optional estimation profile is desktop configuration, not database
 history.
+
+## 14. Training knowledge read boundary
+
+Training Knowledge V1 adds no table, migration, seed data or synchronization
+artifact. The desktop database remains schema v11. Read-only context assembly
+joins an exact existing exercise with its persisted BODY ZONE relations,
+occurrence history, raw sets, actual equipment and latest explicit MAX, then
+optionally attaches immutable catalog knowledge. Missing catalog knowledge is
+valid and never causes a database mutation. The scientific catalog's BODY ZONE
+projection is not stored in `exercise_body_zones` and does not replace user
+classification. See [Training knowledge system V1](domain/knowledge_system.md).
+Occurrence pagination and latest explicit MAX use an exact derived instant
+comparison, then bytewise session/entry ID ties. They scan all matching
+metadata and retain only bounded candidates before hydration, inside a read
+snapshot. Original timestamp text stays unchanged. Malformed storage and a
+selected timestamp beyond the existing 40-character C output capacity produce
+explicit errors. The [temporal contract](reviews/training_knowledge_v1_temporal_contract.md)
+distinguishes the accepted profile from accidental Python ISO extensions.
+The temporal contract has independent PASS evidence. The initial audit's stale
+temporal documentation, Android loader, Meson input, and C role-only query
+findings were repaired, independently verified, and final-validated;
+`TRAINING_KNOWLEDGE_V1=PASS`.

docs/domain/anatomy_and_movement.md

@@ -0,0 +1,113 @@
+# Functional anatomy and movement knowledge
+
+TRAINING KNOWLEDGE V1 separates anatomy from the exercise, its execution and
+its physical equipment. A muscle can move a joint, assist another mover or
+stabilize a segment; its role changes with posture, resistance direction and
+movement phase. The catalogs record qualitative roles, not force percentages,
+effective-set fractions or physiological measurements.
+
+The scientific references are in `catalog/science-references-v1.json`.
+Functional entities, joint actions and authored movement conventions belong in
+`catalog/muscles-v1.json`, `catalog/joint-actions-v1.json` and
+`catalog/movement-patterns-v1.json`. A muscle region or muscle group is not a
+new anatomical muscle. Parent groups and their component muscles must not be
+summed as independent exposure.
+
+## Evidence and confidence
+
+Every interpretation distinguishes established anatomy, biomechanical
+interpretation, EMG evidence, intervention evidence, manufacturer statements
+and practical inference. Anatomy explains plausible function; longitudinal
+training studies address adaptation. Surface EMG measures a signal affected
+by recording and physiological conditions. Greater amplitude does not establish
+greater muscle force, hypertrophy, strength improvement or universal primary
+muscle status. [Vigotsky and colleagues](https://pubmed.ncbi.nlm.nih.gov/29354060/)
+
+Confidence uses exactly `high`, `moderate` and `uncertain`. Established anatomy
+can have high confidence while its application to an unobserved machine variant
+remains uncertain. Moderate confidence is appropriate when the exercise family
+is clear but geometry changes secondary or stabilizing roles. Missing evidence
+is retained explicitly, not replaced with a commercial-name rule.
+
+## Functional muscle coverage
+
+The catalog covers these functional distinctions:
+
+| Area | Functional distinctions |
+|---|---|
+| Chest | Pectoralis major, clavicular and sternocostal regions; pectoralis minor as a scapular muscle |
+| Back and scapula | Latissimus dorsi, teres major, trapezius regions, rhomboids and serratus anterior |
+| Shoulder | Anterior, middle and posterior deltoid; supraspinatus, infraspinatus, teres minor and subscapularis |
+| Arms and forearms | Elbow flexors, triceps, grip/wrist flexors and extensors, pronation and supination |
+| Trunk | Rectus abdominis, obliques, transversus abdominis, erector spinae, multifidus and quadratus lumborum |
+| Hip | Gluteal muscles, tensor fasciae latae, iliopsoas and adductors |
+| Thigh | Quadriceps with biarticular rectus femoris distinguished from vasti; biarticular hamstrings distinguished from biceps femoris short head |
+| Lower leg | Gastrocnemius, soleus and tibialis anterior |
+
+Upper-limb anatomy supports shoulder, scapular, elbow and forearm distinctions.
+Scapular movement is not interchangeable with glenohumeral movement. Stabilizing
+the humeral head is not the same task as dynamically rotating the shoulder.
+[OpenStax upper-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-5-muscles-of-the-pectoral-girdle-and-upper-limbs)
+
+Pectoral regions share actions but differ in orientation and contribution across
+shoulder positions. Their existence does not establish separate isolatable
+“upper” and “lower” chest muscles. Deltoid regions likewise have different
+lines of action; the movement identifies the likely emphasis.
+[NIH pectoralis anatomy](https://www.ncbi.nlm.nih.gov/books/NBK525991/),
+[NIH deltoid anatomy](https://www.ncbi.nlm.nih.gov/books/NBK537056/)
+
+The hip and knee distinctions are essential: knee extension and knee flexion
+are different functions despite both mapping to `thighs`. Gastrocnemius crosses
+the knee and ankle; soleus does not cross the knee. Hip flexion changes the
+length of biarticular hamstrings but not biceps femoris short head.
+[OpenStax lower-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-6-appendicular-muscles-of-the-pelvic-girdle-and-lower-limbs),
+[Maeo and colleagues](https://pubmed.ncbi.nlm.nih.gov/33009197/)
+
+Gluteus minimus abducts and stabilizes the hip. Some lower-limb textbook figure
+alternative text describes direction inconsistently, including minimus and
+adductor examples; those descriptions are not copied as anatomical truth.
+The NIH account corroborates the minimus classification.
+[NIH gluteus minimus anatomy](https://www.ncbi.nlm.nih.gov/books/NBK556144/)
+
+Spinal flexion/extension and hip flexion/extension must remain separate.
+Abdominal-wall muscles combine movement and tension/control functions, while
+posterior spinal muscles contribute extension and segmental control.
+[OpenStax spinal anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-3-axial-muscles-of-the-head-neck-and-back),
+[OpenStax trunk anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-4-axial-muscles-of-the-abdominal-wall-and-thorax),
+[NIH abdominal-wall anatomy](https://www.ncbi.nlm.nih.gov/books/NBK525975/)
+
+## Actions, patterns and stabilization
+
+Joint actions describe motion. Their definitions include shoulder and scapular
+actions, elbow flexion/extension, forearm rotation, hip actions, knee actions,
+ankle actions and trunk flexion/extension/rotation/lateral flexion.
+A loaded return may reverse the visible joint motion while the same agonists
+control it eccentrically.
+[OpenStax movement terminology](https://openstax.org/books/anatomy-and-physiology-2e/pages/9-5-types-of-body-movements)
+
+Horizontal/vertical push and pull, knee dominant, hip dominant and single-joint
+patterns are Trainlog programming conventions grounded in those actions.
+They are not universally standardized anatomical categories or exact torque
+ratios. A single-joint exercise can involve many muscles and stabilizers.
+A dip grouped as a vertical push is still mechanically different from an
+overhead press.
+
+Trunk stabilization describes resisting an external moment while limiting
+motion. Anti-extension, anti-rotation and anti-lateral-flexion are task demands,
+not invented joint movements. Locomotion, cyclic pedaling and cyclic rowing
+remain distinct; a cardio profile alone does not identify the action sequence.
+
+## BODY ZONES projection
+
+`catalog/body-zones-v1.json` remains the authoritative UX taxonomy. Its zones
+are coarser than functional anatomy. Forearm muscles fall under `arms`,
+posterior trunk muscles can relate to both `back` and `core`, and the lateral
+thorax/scapular function of serratus anterior does not fit a simple surface
+location rule. Scientific muscle-level projections explain these conventions;
+they do not alter persisted exercise relations.
+
+A secondary BODY ZONE need not list every accessory or stabilizing muscle.
+Absence of `shoulders` on a row does not deny posterior-deltoid participation.
+Absence of `thighs` on hip abduction does not deny tensor fasciae latae
+participation. `full_body` is not an automatic synonym for cardio or a command
+to mark every zone.

docs/domain/exercise_equipment_interpretation.md

@@ -0,0 +1,117 @@
+# Exercise and equipment interpretation
+
+Scientific knowledge is keyed to existing stable exercise and equipment
+identities. Display labels are identification clues. They are not anatomical
+proof, runtime classification rules or grounds for merging identities.
+`catalog/exercise-knowledge-v1.json` separates reviewed interpretations from
+conditional candidates; `catalog/equipment-knowledge-v1.json` describes
+physical apparatus and its exercise capabilities.
+
+The reviewed inventory contains 23 exercise identities and 41 equipment
+identities: 38 supplied definitions and three durable custom entries. No
+manufacturer or model has been confirmed for the local apparatus. Available
+manufacturer examples remain examples; their ratios, trajectories and outcome
+claims are not transferred to the local inventory.
+
+## Reviewed exercise families
+
+The following are biomechanical interpretations using the stated execution.
+They are qualitative classifications, not measurements of individual force
+contributions. Anatomy and the references linked below support the rationale.
+
+| Actual catalog exercise | Interpretation | Existing primary / secondary BODY ZONES | Main qualification |
+|---|---|---|---|
+| Abdominal crunch | Resisted spinal flexion; rectus and obliques | core | Confirm that movement is not predominantly hip flexion |
+| Arm curl | Elbow flexion; biceps, brachialis and brachioradialis | arms | Grip and shoulder support change participation |
+| Back extension | Spinal extension; erector spinae and multifidus | back / core | Pelvic restraint determines hip involvement |
+| Converging Shoulder Press | Overhead push; deltoids with elbow extension | shoulders / arms | Plane, seat support and scapular freedom unresolved |
+| Diverged seated row; Seated row | Horizontal pull; humeral extension and scapular retraction | back / arms | Elbow path determines lat/scapular/posterior-deltoid emphasis |
+| Diverging lat pulldown; Lat pull | Vertical pull; shoulder adduction/extension and elbow flexion | back / arms | Grip and linkage do not establish regional isolation |
+| Hip abduction | Abduction involving medius/minimus and other abductors | glutes | Tensor fasciae latae and superior maximus contributions vary |
+| Hip adduction | Resisted thigh approximation by hip adductors | thighs | Hip and knee angles affect individual muscles |
+| Leg extension | Resisted knee extension by quadriceps | thighs | Rectus femoris also crosses hip |
+| Leg press | Combined knee/hip extension | thighs / glutes | Machine, depth and foot placement alter moments |
+| Prone leg curl | Knee flexion with prone support | thighs | Distinct length context from seated curl |
+| Seated Leg; Seated leg curl | Knee flexion with seated support | thighs | Two existing identities retained; common family does not merge IDs |
+| Rear Delt | Reverse fly: humeral horizontal abduction with scapular contribution | shoulders / back | Distinct from Pec Fly on the same device |
+| Rotary torso | Relative thorax-pelvis rotation; paired oblique action | core | Compatible supplied equipment is not proven occurrence linkage |
+
+Chest Press priority is addressed through a complete **conditional** anterior
+press interpretation: pectoralis major as a primary mover, anterior deltoid and
+triceps as synergists, shoulder horizontal adduction with elbow extension,
+and `horizontal_push`. The custom `Chest press` and `Converting chest press`
+identities have not independently established that execution. Their retained
+chest/shoulders/arms relations are plausible historical mappings; the scientific
+candidate stays uncertain until apparatus and motion are confirmed.
+[NIH pectoralis anatomy](https://www.ncbi.nlm.nih.gov/books/NBK525991/),
+[OpenStax upper-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-5-muscles-of-the-pectoral-girdle-and-upper-limbs)
+
+The custom `Abdominal` entry similarly retains a conditional crunch candidate,
+not an asserted execution. Both generic warm-up identities remain unresolved.
+`Marche` has a conditional walking interpretation; a continuous/duration profile
+and an aggregate treadmill inventory do not establish all execution details or
+a specific occurrence association. Unknown new custom exercises remain
+unclassified until evidence is added explicitly.
+
+Seated versus prone curl has direct longitudinal evidence: the studied seated
+condition produced greater growth of biarticular hamstrings. This supports
+retaining length context while avoiding a universal machine or outcome ranking.
+[Maeo and colleagues](https://pubmed.ncbi.nlm.nih.gov/33009197/)
+
+The leg-press review concerns EMG. It supports quadriceps involvement but does
+not establish a robust universal foot-placement recipe or prove that hamstring
+coactivation replaces knee-flexion training.
+[Martín-Fuentes and colleagues](https://pubmed.ncbi.nlm.nih.gov/32605065/)
+
+Reverse pec-deck and row/pulldown investigations concern selected EMG tasks.
+They corroborate plausible posterior-deltoid and scapular contributions without
+ranking long-term growth. Divergent/convergent machine labels add no independent
+outcome evidence.
+[Franke and colleagues](https://pubmed.ncbi.nlm.nih.gov/24947920/),
+[Lehman and colleagues](https://pubmed.ncbi.nlm.nih.gov/15228624/)
+
+Pelvic stabilization changes lumbar-extensor excitation. A Back Extension
+execution dominated by hip motion may deserve different muscle roles from the
+spinal-extension interpretation; that decision requires actual geometry and
+execution evidence.
+[Lee](https://pubmed.ncbi.nlm.nih.gov/26730390/)
+
+## Multifunction and unrestricted equipment
+
+A Rear Delt / Pec Fly device supports at least two distinct tasks. Rear Delt
+uses shoulder horizontal abduction and a shoulder/back projection. Pec Fly
+uses horizontal adduction, with a chest emphasis and possible anterior-deltoid
+assistance. Pec Fly has no verified local exercise ID. It remains an unlinked
+capability, not a fabricated catalog exercise.
+
+The assistance device likewise supports separate assisted dip and assisted
+chin/pull-up capabilities. Neither has a verified local exercise ID. The
+legacy `assisted_dip` and `assisted_chin` strings in the equipment manifest are
+not creator exercise identities. Chin-up and pronated pull-up also require
+specific grip/execution context. Manufacturer catalogs demonstrate that such
+combination products exist, not that these examples identify the local model.
+[Life Fitness equipment catalogue](https://www.lifefitness.com.au/wp-content/uploads/2024/02/Life-Fitness-Catalogue_2024_web.pdf)
+
+Functional trainers, adjustable pulleys, dumbbells, kettlebells, bags,
+medicine balls and suspension straps require the actual exercise. Their
+presence cannot establish a primary BODY ZONE. Cable height and routing,
+attachment, stance and line of pull determine the task. A documented example
+with a 4:1 cable ratio illustrates why local ratios must be verified separately.
+[Precor RUD0915](https://www.precor.com/en-US/products/RUD0915)
+
+Benches and guided squat machines require support and movement details. Cardio
+apparatus requires mode, speed/cadence, resistance and support context. A rowing
+ergometer includes a leg/trunk/arm cycle and is not the same exercise as a
+seated resistance row. Battle ropes have mass and inertia: their existing
+`bodyweight` catalog category is retained as legacy metadata, not a calibrated
+physical load model. Any future vocabulary change needs a separate explicit
+compatibility decision.
+
+## BODY ZONE audit result
+
+No reviewed finding justifies an automatic persisted-zone change. Existing
+assignments are compatible with the identified families or remain conditional
+where observation is missing. Additional potential synergists are documented
+at muscle level. A future correction must state its evidence, affected stable
+identities and explicit migration separately; enrichment never silently
+rewrites historical mappings.

docs/domain/knowledge_audit.md

@@ -0,0 +1,29 @@
+# Training knowledge science audit
+
+Generated deterministically from the six canonical knowledge catalogs. Conditional rows are candidates that require explicit confirmation and do not participate in ordinary resolved queries.
+
+| ID | Name | Actions | Patterns | Primary | Secondary | Stabilizers | Science primary zone | Science secondary zones | Existing primary zone | Existing secondary zones | Equipment | Confidence | Source refs | Audit status |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| ex_01ff06dd-ad00-46ee-9b46-ed32cabedbef | Hip adduction | hip_adduction | single_joint_hip_adduction | hip_adductors | — | — | thighs | — | thighs | — | hip_adduction | high | openstax_lower | confirmed |
+| ex_1872246a-39ae-44dc-b58d-f87e90ca49ab | Leg extension | knee_extension | single_joint_knee_extension | quadriceps | — | — | thighs | — | thighs | — | leg_extension | high | openstax_lower | confirmed |
+| ex_1a34814c-2e46-40fc-b1f4-6d60b8e5a3e0 | Rotary torso | trunk_rotation | trunk_rotation | external_oblique, internal_oblique | multifidus | erector_spinae, transversus_abdominis | core | — | core | — | rotary_torso | moderate | nih_abdominal_wall, openstax_back, openstax_trunk | confirmed |
+| ex_1b0c6b8b-b05e-4e6f-8809-5f7d85d668de | Diverged seated row | elbow_flexion, scapular_retraction, shoulder_extension, shoulder_horizontal_abduction | horizontal_pull | latissimus_dorsi, rhomboids, trapezius_middle | biceps_brachii, brachialis, brachioradialis, deltoid_posterior, teres_major | erector_spinae, forearm_extensors, forearm_flexors | back | arms | back | arms | diverging_seated_row | moderate | franke_2015, lehman_2004, openstax_upper | confirmed |
+| ex_2d488c08-194c-4051-a3c9-34471646c1d3 | Gym échauffement | — | — | — | — | — | — | — | — | — | — | uncertain | — | unresolved |
+| ex_33f79331-871c-4eed-babe-346e53a99070 | Seated row | elbow_flexion, scapular_retraction, shoulder_extension, shoulder_horizontal_abduction | horizontal_pull | latissimus_dorsi, rhomboids, trapezius_middle | biceps_brachii, brachialis, brachioradialis, deltoid_posterior, teres_major | erector_spinae, forearm_extensors, forearm_flexors | back | arms | back | arms | seated_row | moderate | franke_2015, lehman_2004, openstax_upper | confirmed |
+| ex_34a5c9c3-c032-4dfb-be32-8bf09832f72b | Converting chest press | elbow_extension, shoulder_horizontal_adduction | horizontal_push | pectoralis_major | deltoid_anterior, triceps_brachii | infraspinatus, subscapularis, supraspinatus, teres_minor | chest | arms, shoulders | chest | arms, shoulders | eq_c660cd61-5b17-498e-8d51-9eafcf7e2731 | uncertain | nih_deltoid, nih_pectoralis, openstax_upper | questionable |
+| ex_43c7375f-934c-4650-930c-45807d2f2929 | Gym/Échauffement | — | — | — | — | — | — | — | — | — | — | uncertain | — | unresolved |
+| ex_474ec393-3efa-4aaa-8e08-1a0245ed7835 | Back extension | trunk_extension | trunk_extension | erector_spinae | multifidus | external_oblique, internal_oblique, rectus_abdominis | back | core | back | core | back_extension | moderate | lee_2015, openstax_back, openstax_lower | questionable |
+| ex_4bd03d55-e644-436e-876e-07837824bde5 | Abdominal | trunk_flexion | trunk_flexion | rectus_abdominis | external_oblique, internal_oblique | transversus_abdominis | core | — | core | — | eq_3f987a36-bbb4-4e29-9c9f-1f201f1596e0 | uncertain | nih_abdominal_wall, openstax_trunk | questionable |
+| ex_4cd2433e-80b1-478a-b8df-73fc6ef80962 | Rear Delt | scapular_retraction, shoulder_horizontal_abduction | single_joint_shoulder_horizontal_abduction | deltoid_posterior | rhomboids, trapezius_middle | infraspinatus, subscapularis, supraspinatus, teres_minor | shoulders | back | shoulders | back | rear_delt_pec_fly | moderate | franke_2015, nih_deltoid, openstax_upper, vigotsky_2018 | confirmed |
+| ex_617007f9-7420-4408-91b9-8ffb77900f13 | Seated Leg | knee_flexion | single_joint_knee_flexion | hamstrings | gastrocnemius | — | thighs | — | thighs | — | seated_leg_curl | high | maeo_2021, openstax_lower | confirmed |
+| ex_6dfc7ffd-8891-464e-a995-808baf1b0d7b | Converging Shoulder Press | elbow_extension, scapular_upward_rotation, shoulder_abduction, shoulder_flexion | vertical_push | deltoid_anterior, deltoid_middle | serratus_anterior, trapezius_lower, trapezius_upper, triceps_brachii | infraspinatus, subscapularis, supraspinatus, teres_minor | shoulders | arms | shoulders | arms | converging_shoulder_press | moderate | nih_deltoid, openstax_upper | confirmed |
+| ex_7e7cf906-2214-4066-bcb7-c16382d83b3b | Hip abduction | hip_abduction | single_joint_hip_abduction | gluteus_medius, gluteus_minimus | gluteus_maximus, tensor_fasciae_latae | — | glutes | — | glutes | — | hip_abduction | moderate | nih_gluteus_minimus, openstax_lower | confirmed |
+| ex_8552dd77-fcd7-4f06-a1cc-d956eb1009af | Chest press | elbow_extension, shoulder_horizontal_adduction | horizontal_push | pectoralis_major | deltoid_anterior, triceps_brachii | infraspinatus, subscapularis, supraspinatus, teres_minor | chest | arms, shoulders | chest | arms, shoulders | eq_0a462c1e-9fb6-4fd7-b0a2-53a0e86f33c6 | uncertain | nih_deltoid, nih_pectoralis, openstax_upper | questionable |
+| ex_9adf7566-f10c-443c-b63b-681d37665693 | Abdominal crunch | trunk_flexion | trunk_flexion | rectus_abdominis | external_oblique, internal_oblique | transversus_abdominis | core | — | core | — | abdominal | moderate | nih_abdominal_wall, openstax_trunk | confirmed |
+| ex_a1ef5047-b44b-4c64-a6ed-c7a3bc13b163 | Seated leg curl | knee_flexion | single_joint_knee_flexion | hamstrings | gastrocnemius | — | thighs | — | thighs | — | seated_leg_curl | high | maeo_2021, openstax_lower | confirmed |
+| ex_a72fa713-4b0e-431d-95e2-42d95beb77b1 | Lat pull | elbow_flexion, scapular_downward_rotation, shoulder_adduction, shoulder_extension | vertical_pull | latissimus_dorsi | biceps_brachii, brachialis, brachioradialis, teres_major | forearm_extensors, forearm_flexors, infraspinatus, subscapularis, supraspinatus, teres_minor | back | arms | back | arms | lat_pull | moderate | lehman_2004, openstax_upper, vigotsky_2018 | confirmed |
+| ex_b1e6ffc6-75b5-45ff-a3c0-e7433c58013d | Marche | ankle_dorsiflexion, ankle_plantarflexion, hip_extension, hip_flexion, knee_extension, knee_flexion | locomotion | gastrocnemius, gluteus_maximus, quadriceps, soleus | hamstrings, iliopsoas, tibialis_anterior | erector_spinae, gluteus_medius, gluteus_minimus | full_body | — | — | — | — | uncertain | nih_gluteus_minimus, openstax_lower | unresolved |
+| ex_b432623f-bfe9-4daf-a653-60ec7fdffbde | Leg press | hip_extension, knee_extension | knee_dominant | quadriceps | adductor_magnus, gluteus_maximus | gastrocnemius, soleus | thighs | glutes | thighs | glutes | leg_press, plate_loaded_leg_press | moderate | martin_fuentes_2020, openstax_lower, vigotsky_2018 | confirmed |
+| ex_b4d1daf1-de4a-4016-abdf-487bf6014ce6 | Diverging lat pulldown | elbow_flexion, scapular_downward_rotation, shoulder_adduction, shoulder_extension | vertical_pull | latissimus_dorsi | biceps_brachii, brachialis, brachioradialis, teres_major | forearm_extensors, forearm_flexors, infraspinatus, subscapularis, supraspinatus, teres_minor | back | arms | back | arms | diverging_lat_pulldown | moderate | lehman_2004, openstax_upper, vigotsky_2018 | confirmed |
+| ex_d7398d9f-d928-4d2e-94e9-74e201da55c5 | Prone leg curl | knee_flexion | single_joint_knee_flexion | hamstrings | gastrocnemius | — | thighs | — | thighs | — | prone_leg_curl | high | maeo_2021, openstax_lower | confirmed |
+| ex_ec619fc2-4685-4044-873c-86764bd4a0fe | Arm curl | elbow_flexion | single_joint_elbow_flexion | biceps_brachii, brachialis | brachioradialis | forearm_extensors, forearm_flexors | arms | — | arms | — | arm_curl | high | openstax_upper | confirmed |

docs/domain/knowledge_system.md

@@ -0,0 +1,129 @@
+# Training knowledge system V1
+
+`TRAINING_KNOWLEDGE_V1` is an implemented read-only, evidence-linked knowledge
+layer with lifecycle status `TRAINING_KNOWLEDGE_V1=PASS`. Its bounded scientific
+review, independent temporal review, final engineering audit, repair
+verification, and final executable validation passed. It does not change the frozen
+exchange formats, either SQLite schema, synchronization,
+or the meaning of an exercise, BODY ZONE, occurrence, set, or measured maximum.
+
+The authored scientific source is the six versioned JSON catalogs in
+[`catalog/`](../../catalog/):
+
+- `science-references-v1.json` (23 references);
+- `muscles-v1.json` (53 muscles);
+- `joint-actions-v1.json` (35 joint actions);
+- `movement-patterns-v1.json` (26 movement patterns);
+- `exercise-knowledge-v1.json` (23 exercise records);
+- `equipment-knowledge-v1.json` (41 equipment records).
+
+The catalog loader and validator enforce IDs, ordering and cross-references.
+The C representation is generated from those assets; Android loads the same
+assets through `TrainingKnowledgeCatalog`. Neither C nor Kotlin contains a
+manually maintained duplicate scientific table. The generated knowledge audit
+is an audit artifact and is not an authored source or a file to edit directly;
+the retained review-oriented [knowledge audit](knowledge_audit.md) and
+[`training-knowledge-audit-v1.json`](../../catalog/training-knowledge-audit-v1.json)
+provide the current navigable audit records.
+
+The catalog contains no runtime history and does not seed either database.
+Unknown future exercise and equipment IDs remain valid runtime data. A display
+name, legacy slug, equipment family, or catalog label never substitutes for an
+actual `ex_<uuid-v4>` exercise ID. Consequently the documented capabilities
+for Pec Fly, Assisted Dip and Assisted Chin do not create associations until a
+real runtime exercise UUID exists. The two Leg press equipment variants remain
+separate contexts; the unobserved Rotary compatibility never enters history.
+
+## Scientific scope and uncertainty
+
+Scientific mappings describe anatomy, mechanics, movement patterns and the
+BODY ZONE projection used by the knowledge catalog. They do not overwrite the
+persisted primary/secondary BODY ZONE relations selected for a runtime
+exercise. Source type, notes, limitations, confidence and conditional
+interpretation remain available through the APIs.
+
+The current inventory has six high-confidence, eleven moderate-confidence and
+six uncertain exercise records. It represents 20 initial zone-catalog exercise
+IDs plus three further reviewed IDs, 38 supplied equipment IDs and three
+observed custom IDs. The BODY ZONE audit has 16 confirmed, four questionable
+and three unresolved records. It records no database mutation. `Marche` is a
+conditional candidate and remains excluded by normal resolved-knowledge
+filters; two warmups and the custom Chest press, Converting and Abdominal
+entries remain unresolved until their execution is known. Across equipment,
+21 records are scientifically documented, 17 are mechanically identified with
+incomplete anatomy, and three have uncertain equipment identity. All 41 have
+unknown manufacturer and model: a generic-family source does not prove a local
+machine model.
+
+The detailed evidence and limitations are in
+[anatomy and movement](anatomy_and_movement.md),
+[exercise and equipment interpretation](exercise_equipment_interpretation.md),
+[programming foundations](programming_foundations.md), and the central
+[`science-references-v1.json`](../../catalog/science-references-v1.json).
+Project domain decisions follow the
+[`trainlog-anatomy` guidance](../../.agents/skills/trainlog-anatomy/SKILL.md).
+
+## Read-only application contracts
+
+On desktop, [`training_knowledge.h`](../../tui/include/trainlog/training_knowledge.h)
+provides immutable lookup, enumeration and resolved-knowledge query APIs. All
+strings and records are borrowed generated storage with process-lifetime
+validity; stable IDs, rather than labels, are keys. The query AND-combines its
+optional scientific-zone, movement-pattern, muscle-role and available-equipment
+filters. Conditional and unresolved entries cannot match it.
+
+[`training_context.h`](../../tui/include/trainlog/training_context.h) composes
+one exact runtime `exercise_id` with its persisted zones, optional scientific
+record, compatible equipment, latest explicit MAX, and actual occurrence/set
+history. Missing science is valid; it never fabricates history. One load uses a
+nested-safe read snapshot. It accepts an occurrence limit of 1–32 and a set
+preview limit of 1–64 per occurrence; empty history remains empty. Follow-up cursors order current data by the
+exact represented instant, then bytewise session ID and occurrence ID; they are not a
+snapshot across calls, including when equivalent instants use different
+offset text.
+
+The corrected C and Android readers use one explicit parser/comparator policy
+for stored timestamps and exclusive cursors. They admit extended dates,
+`T/t`, `Z/z`, offsets through `23:59`, arbitrary exact dot fractions and the
+Android writer's omitted-seconds form. They scan all matching metadata and
+retain bounded winners before hydration; no SQLite date parser selects the
+page. Source text and identities remain unchanged. Malformed timestamps fail
+explicitly, as does a selected value beyond C's existing output capacity.
+The [temporal correction record](../reviews/training_knowledge_v1_temporal_contract.md)
+documents parser-only extensions, limits and passing production regressions.
+
+Android exposes the same scientific lookups through `TrainingKnowledgeCatalog`
+and composes the same boundary through
+`TrainlogRepository.getTrainingExerciseContext()`,
+`listExerciseOccurrences()` and `listExerciseOccurrenceSets()`. Its cursors
+have the same current-data, chronological contract. Returned runtime context
+keeps persisted zone IDs separate from the scientific mapping, includes actual
+equipment and the latest explicit MAX, and preserves raw per-set values.
+
+No API infers range of motion, setup, actual force, or comparability of raw
+kilogram labels. Equipment, resistance semantics and execution context remain
+explicit. There are no prescriptions in V1.
+
+## Documented future pipeline only
+
+The following is an architectural boundary for later work, not a V1 generator,
+scoring algorithm, proposal, database change, or user-interface behavior.
+
+```text
+Session inputs
+  BODY ZONE, duration, goal, available equipment, recent history
+    -> movement functions
+    -> real resolved candidates
+    -> explicit availability
+    -> recent history and explicit-MAX context
+    -> future fatigue/recent-coverage interpretation
+    -> a future proposal
+```
+
+Candidate selection must seek diverse movement patterns rather than repeatedly
+selecting the same muscle. A future multi-session program would additionally
+take the goal, available days, functions/zones, recent frequency, recovery,
+progression, equipment and history, then reason across sessions. Medical
+constraints would be explicit inputs; no diagnosis or rehabilitation status is
+to be inferred. V1 prescribes no exercises, weights, sets, fatigue scores,
+progression or schedule.

docs/domain/programming_foundations.md

@@ -0,0 +1,93 @@
+# Programming foundations and MAX context
+
+TRAINING KNOWLEDGE V1 supplies knowledge for future reasoning about training.
+It does not generate workouts, prescribe loads or provide clinical
+rehabilitation advice. Principles below distinguish intervention findings from
+Trainlog's practical interpretation of them.
+
+## Adaptation and exposure
+
+Resistance-training interventions improve strength and muscle size across many
+studied configurations. Higher loads tend to favor high-load strength outcomes;
+multiple sets characterize better-ranked hypertrophy configurations in a large
+network meta-analysis. These are population-level comparisons, not an optimal
+program for every person. Load, sets and frequency interact.
+[Currier and colleagues](https://pubmed.ncbi.nlm.nih.gov/37414459/)
+
+Volume can mean sets, repetitions or accumulated load. Those are different
+quantities. Machine kilograms multiplied by repetitions do not establish
+equal muscle exposure across different devices. Direct and indirect muscle
+participation cannot be assigned universal equal set credit. Frequency helps
+distribute exposure and fatigue; no weekly frequency or unbounded more-volume
+rule is encoded here.
+
+Relative external load and effort are also different. A heavy task need not
+end in failure; a lighter task can. Evidence does not establish that momentary
+failure is generally required for strength or hypertrophy, but it does not
+imply that arbitrarily easy sets provide the same stimulus. Failure definitions,
+actual effort and compared volume constrain interpretation.
+[Grgic and colleagues](https://pubmed.ncbi.nlm.nih.gov/33497853/),
+[Refalo and colleagues](https://pubmed.ncbi.nlm.nih.gov/36334240/)
+
+Failure conditions generally produce greater acute fatigue in the reviewed
+experiments. Acute markers cannot supply a universal recovery clock or predict
+an individual's long-term result. Repeated performance, actual execution and
+context matter when a future system interprets recovery.
+[Vieira and colleagues](https://pubmed.ncbi.nlm.nih.gov/34881412/)
+
+## Progression, selection and coverage
+
+Progression and specificity require consistent definitions of the task and
+its successful performance. The historical ACSM position stand supplies these
+organizing principles; its numeric schedules and load recommendations are not
+adopted as V1 rules. Comparative outcomes are interpreted using the subsequent
+reviews above.
+[ACSM progression position stand](https://pubmed.ncbi.nlm.nih.gov/19204579/)
+
+Exercise variation can change regional exposure and task practice. The limited
+intervention literature supports considering purposeful variation, not a
+requirement to randomize exercises or change them as often as possible. Stable
+exercise context also makes progress interpretable.
+[Kassiano and colleagues](https://pubmed.ncbi.nlm.nih.gov/35438660/)
+
+Practical selection should compare the intended joint action, muscle role,
+range of motion, stability demand and available equipment. BODY ZONES alone
+cannot establish a substitute. Leg extension and leg curl share `thighs` while
+training opposing knee actions. Seated and prone curls share knee flexion but
+differ in the length context of biarticular muscles.
+[OpenStax lower-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-6-appendicular-muscles-of-the-pelvic-girdle-and-lower-limbs),
+[Maeo and colleagues](https://pubmed.ncbi.nlm.nih.gov/33009197/)
+
+Coverage therefore considers movement functions as well as zones: horizontal
+and vertical pushing/pulling, knee and hip extension, knee flexion, hip
+abduction/adduction, plantarflexion, trunk motion and resisting trunk motion.
+This is an authored reasoning aid. It is not a universal checklist that proves
+a program balanced, effective or suitable for a particular individual.
+
+## What a recorded MAX can establish
+
+`max_weight_kg` records an observed result owned by an exercise occurrence.
+Interpreting it requires its resistance and execution context. At minimum,
+consider exercise identity, physical equipment, external-versus-assistance
+mode, machine settings, range of motion, support, technique and success
+criterion. Unknown context stays unknown.
+
+An explicit MAX result without repetition count is not proven to be a one-rep
+maximum. Ordinary sets do not become measured maxima by being the heaviest
+available observation. No estimated 1RM or percentage prescription follows
+from the value alone. A shared BODY ZONE, movement family or machine does not
+make distinct exercises' results interchangeable.
+
+Stack labels need not equal handle resistance. Pulley routing, lever arms,
+cams, tare and friction affect external moments; human moment arms add further
+variation. Manufacturer documentation for an identified example demonstrates
+a non-unit cable ratio, but supplies no conversion for the unconfirmed local
+equipment.
+[Precor pulley documentation](https://www.precor.com/en-US/products/RUD0915)
+
+For otherwise matched assisted performance, less help represents greater
+unsupported demand. Body mass, assistance mechanism and motion remain part of
+the comparison. Do not treat assistance as added external weight or assume
+that body mass minus displayed assistance gives a calibrated effective load.
+The source exercise and resistance context must remain visible when knowledge
+is composed with history.

docs/exercise_data_model.md

@@ -317,3 +317,30 @@ The newest successful explicit test is the current measured result. A separate
 same-mode historical record may be older.
 
 No estimated 1RM is mixed into this contract.
+
+## 12. Read-only training knowledge context
+
+Training Knowledge V1 adds no exercise profile field and no persisted training
+prescription. It can associate a real stable `exercise_id` with source-linked
+scientific knowledge, compatible equipment and a separate scientific BODY ZONE
+projection. That projection is not the persisted primary/secondary BODY ZONE
+classification and cannot rewrite it.
+
+The composed runtime view contains the actual exercise profile, persisted
+zones, compatible equipment, latest explicit MAX and bounded chronological
+occurrences with raw set values. Unknown custom and future runtime IDs remain
+valid and may have no knowledge record. A name, legacy slug or generic machine
+capability is never made into a phantom UUID association. Raw kilogram labels,
+ROM, setup and actual force retain their recorded/unknown context; they do not
+become interchangeable performance measurements.
+
+See [Training knowledge system V1](domain/knowledge_system.md) for catalog
+ownership, confidence and the documented-only future planning boundary.
+The current implementation lifecycle is `TRAINING_KNOWLEDGE_V1=PASS`. Occurrence pagination and latest
+explicit MAX use the settled shared temporal policy: the admitted source text
+is preserved, instants and arbitrary dot fractions compare exactly, and equal
+instants use bytewise session and entry ID ties. Cursors are exclusive and use
+the same comparison; malformed cursors or matching stored timestamps report
+explicit errors. The initial audit's stale temporal documentation, Android
+loader, Meson input, and C role-only query findings were repaired and
+independently verified; final validation passed.

docs/reviews/training_knowledge_v1_engineering_review.md

@@ -0,0 +1,122 @@
+# Training Knowledge V1 — engineering checkpoint
+
+Date: 2026-09-09. Status: `TRAINING_KNOWLEDGE_V1_ENGINEERING_REVIEW=PASS`.
+
+Scientific review separately passed within its documented scope and uncertainty.
+The independent temporal delta review returned `TEMPORAL_DELTA_REVIEW=PASS`
+with no findings. One configured final-reviewer completed an initial full-tranche
+audit, reported the findings recorded below, and its authorized bounded repair
+chain completed. Independent bounded verification returned
+`FINAL_REVIEW_REPAIR_VERIFICATION=PASS`; fresh final executable validation also
+passed. `TRAINING_KNOWLEDGE_V1=PASS`.
+
+## Current implementation and contract
+
+Six canonical scientific catalogs feed generated C data and the Android asset
+loader. Both clients expose scientific lookup/filter APIs and runtime context
+composition. Android has a collapsible knowledge section; the TUI has a UTF-8
+cell-aware scrolling knowledge screen tested at 72×20. No schema, scientific
+catalog, frozen exchange artifact or stored user history changed in this resume.
+
+The temporal repair covers the two new C readers and their Android counterparts:
+occurrence pagination and latest explicit MAX. The established operational
+profile and parser-only extension classification are recorded in the
+[temporal contract](training_knowledge_v1_temporal_contract.md). All use exact
+integer-second/fraction comparisons and bytewise stable-ID ties. Pagination
+scans all matching metadata, retains at most `limit + 1`, then hydrates only
+selected rows under one read snapshot. Emitted cursors preserve source text.
+
+C has the existing 40-character timestamp output limit: long valid candidates
+are compared exactly, and a selected unrepresentable result fails explicitly.
+Invalid caller cursors and malformed stored timestamps have distinct error
+paths. No malformed date can disappear through SQLite NULL comparison.
+
+Python validation uses the same explicit grammar and exact chronology. The
+parent reproduced and repaired an optional-checker dependency issue after the
+worker implementation: a strict JSON Schema checker could reject omitted
+seconds before Trainlog validation. A private per-call override now aligns the
+schema-format check while preserving all other checker configuration. Actual
+validation-path tests cover absent, strict and permissive optional checkers.
+
+## Original defect retained as historical evidence
+
+The original implementation combined an offset-limited cursor guard with
+SQLite `julianday()` and a different Android parser. Its production probe was:
+
+| Newer source timestamp | Original outcome | Current independent probe |
+|---|---|---|
+| `2026-09-10T10:00:00+02:00` | control passed | PASS |
+| `2026-09-10T10:00:00+14:30` | emitted cursor rejected | PASS |
+| `2026-09-10T10:00:00+15:00` | newer occurrence omitted | PASS |
+| `2026-09-10t10:00:00Z` | newer occurrence omitted | PASS |
+
+The independent current probe extends this to twelve spellings. Every case
+uses two sessions in a temporary database, page size one, exact source/cursor
+assertions, continuation and exhaustion. Both normal and sanitized builds pass.
+Original evidence remains `/tmp/trainlog-knowledge-validation/cursor-probe.c`
+and `cursor-probe.log`; fresh final probe evidence is under
+`/tmp/trainlog-knowledge-review-resume/`.
+
+## Final validation evidence
+
+| Check | Result |
+|---|---|
+| `meson compile -C build` | PASS, fresh strict warning-as-error build |
+| `meson test -C build --print-errorlogs` | 42 passed |
+| Knowledge validator/generation tests | PASS; eight Python tests |
+| Temporal Python tests | PASS; four tests including optional-checker independence |
+| JSON/import-contract validators | PASS; six import cases |
+| Three standalone public C17 headers | PASS, including `-pedantic-errors` |
+| Two targeted C tests under ASan/UBSan | PASS; timestamp Python test also passes in that test invocation |
+| Independent twelve-form C production probe | PASS normally and under ASan/UBSan |
+| Android unit tests and debug assembly, Java 17 | PASS; 56 tests, zero failures/errors, one known fixture skip |
+| Project skill validator and `git diff --check` | PASS |
+
+The known Android skip is `RealAndroidV9BodyZonesMigrationTest`, because its
+external `TRAINLOG_ANDROID_V9_FIXTURE` is unavailable. No knowledge/temporal
+test is skipped. Manual MTP, installed-device UI and real-terminal visual
+checks were not performed or claimed. Sanitizer evidence is scoped to the
+new affected C paths, not the complete GUI executable.
+
+## Independent review and final-review repair record
+
+The isolated temporal review was independent of the original implementation
+and reviewed grammar/calendar constraints, offset bounds, fractions, byte ties,
+cursor aliasing/exclusivity, source capacity, snapshots, and Android/Python
+parity. It ran Meson `training_context` and `timestamp_validation` plus four
+Python temporal tests, returned `TEMPORAL_DELTA_REVIEW=PASS`, and required no
+repair.
+
+The configured final-reviewer then audited the full tranche. Its initial result
+was FAIL and identified these concrete findings:
+
+| Severity | Finding | Authorized repair recorded in `/tmp/trainlog-knowledge-review-resume/repair.md` |
+|---|---|---|
+| BLOCKER | Stale temporal-defect documentation in `architecture.md`, `exercise_data_model.md`, and `tui.md` | Replaced claims of an unresolved reader defect with the independently passing temporal contract and correct repair/validation state. |
+| BLOCKER | Android loader parity holes | Enforced canonical exercise/equipment identity syntax, reverse capability compatibility, HIGH evidence source type, and non-unresolved BODY ZONE audit evidence; added loader regressions. |
+| BLOCKER | Meson generator inputs omitted `body-zones-v1.json` and `equipment-v1.json` although validation reads them | Declared both files as `training_knowledge_generated` inputs and proved regeneration without output-byte change. |
+| HIGH | C query accepted a muscle role without a muscle ID, unlike Android's paired filter | Rejected role-only queries, documented the paired contract in the public header, and added a C regression. |
+
+The repair report records strict C17 syntax, knowledge validation, generated
+output equality, targeted Meson test, Android loader tests, dependency proof,
+and `git diff --check` as PASS. Independent bounded verification closed all
+four findings and explicitly returned
+`FINAL_REVIEW_REPAIR_VERIFICATION=PASS` and
+`TRAINING_KNOWLEDGE_V1_ENGINEERING_REVIEW=PASS`. The fresh tester matrix then
+passed: strict build; 42 Meson tests; eight knowledge and four temporal Python
+tests; knowledge/JSON/import validators; three strict C17 headers; affected C
+knowledge/context tests under ASan/UBSan plus timestamp validation; normal and
+sanitized 12-form temporal probes; Android 56 tests with zero failures/errors
+and one known missing-fixture skip; Java 17 debug assembly; skill validation;
+and diff check. Fresh generated C is byte-identical with SHA-256
+`e8c099f67eb111d61621b5d76592c049823af5508d43e73ec646f22e4c377fca`; all six
+Android assets are byte-identical.
+
+## Remaining observations
+
+The prior bounded review's secondary scientific BODY ZONE IDs on the TUI
+knowledge page remain a nonblocking presentation inconsistency. Local machine
+model/execution uncertainty and missing real exercise UUIDs remain as recorded
+in the scientific review; no phantom identity or scientific reassessment was
+introduced. Older lexical timestamp readers, arbitrary-length C output and
+verified leap-event support are outside this bounded repair.

docs/reviews/training_knowledge_v1_resume.md

@@ -0,0 +1,120 @@
+# Training Knowledge V1 — continuation record
+
+Checkpoint: 2026-09-09. Status: `TRAINING_KNOWLEDGE_V1=PASS`.
+
+This is a completed closeout record. Do not restart science, temporal
+implementation, broad audit, or the bounded repair chain. The independent
+temporal review returned `TEMPORAL_DELTA_REVIEW=PASS` with no findings. One
+deep final audit initially failed, one bounded repair chain closed its four
+findings, independent verification returned
+`FINAL_REVIEW_REPAIR_VERIFICATION=PASS` and
+`TRAINING_KNOWLEDGE_V1_ENGINEERING_REVIEW=PASS`, and fresh final executable
+validation passed.
+
+## Baseline and protections
+
+- Repository: `/home/fy59/Documents/trainlog`.
+- Baseline HEAD: `1aad7b48b3143783ccfd326c0145f495fd568442`.
+- The initial worktree was clean; the current delta belongs to this tranche.
+- This resume initially matched every file hash in the preserved final manifest.
+- No stage, commit, push, reset, restore, stash or clean was performed.
+- Desktop schema remains v11 and Android schema remains v10.
+- Real database logical contents/counts and all scientific/frozen catalogs are
+  preserved. No device install, uninstall, keystore or user-history change.
+
+## Completed during the temporal correction
+
+1. `advisor` / Terra-high completed the isolated contract analysis. Its retained
+   handoff is `/tmp/trainlog-temporal-contract-settled.md`; the durable repository
+   record is [temporal contract](training_knowledge_v1_temporal_contract.md).
+2. `worker` / Sol-medium implemented exact C/Kotlin timestamp parsers,
+   scan/select/hydrate occurrence and latest-MAX reads, coherent exclusive
+   cursors, nested read snapshots and production regressions. Python validation
+   uses the same grammar and exact fractional comparison.
+3. Parent's independent production C probe passed all twelve temporal spellings
+   through two sessions, page size one, source/cursor equality and exhaustion.
+4. A bounded additional validator defect was reproduced: a strict optional
+   JSON Schema format checker could still reject the Android no-seconds form
+   before semantic validation. A per-validation `date-time` override now uses
+   the Trainlog parser without mutating global/caller checkers. Document-path
+   regressions cover missing, strict and permissive checkers.
+5. Full executable validation passed, including the independent C probe under
+   ASan/UBSan. Canonical documentation describes the settled behavior.
+
+Scientific review remains PASS within its recorded scope; no scientific delta
+was introduced. Its reviewed catalog hashes remain applicable.
+
+## Completed independent review and repair chain
+
+The independent temporal reviewer covered grammar, calendar and offset ranges,
+fractions, byte ties, cursor aliasing/exclusivity, source capacity, snapshots,
+and Android/Python parity. It ran the targeted Meson and four Python temporal
+tests and returned `TEMPORAL_DELTA_REVIEW=PASS`, with no temporal repair.
+
+The one final-reviewer audited the complete tranche. Its initial FAIL found
+stale temporal-defect documentation (BLOCKER), Android loader parity holes
+(BLOCKER), missing Meson generator dependencies (BLOCKER), and C acceptance of
+a role-only muscle query (HIGH). The authorized repair report is
+`/tmp/trainlog-knowledge-review-resume/repair.md`: it records the exact loader
+constraints, Meson input declarations, C paired-filter rejection, regressions,
+and repair-scope validation. Independent review of that delta closed all four
+findings in `/tmp/trainlog-knowledge-review-resume/repair-verification.md`.
+No scientific catalog, schema, frozen format, temporal implementation, or
+user-history content changed.
+
+## Closeout evidence
+
+Fresh final validation recorded in
+`/tmp/trainlog-knowledge-review-resume/full-results.json`,
+`android-results.json`, `sanitizer-results.json`,
+`probe-normal-results.json`, and `android-counts.json` passed: strict build;
+42 Meson tests; eight knowledge and four temporal Python tests;
+knowledge/JSON/import validators; three strict C17 headers; affected C
+knowledge/context tests under ASan/UBSan plus timestamp validation; normal and
+sanitized independent 12-form temporal probes; Android 56 tests with zero
+failures/errors and one known missing-real-v9-fixture skip; Java 17 debug
+assembly; skill validation; and diff check. Fresh generated C is byte-identical
+with SHA-256 `e8c099f67eb111d61621b5d76592c049823af5508d43e73ec646f22e4c377fca`;
+all six Android assets are byte-identical.
+
+Preservation before and after repair confirms desktop schema v11 and Android
+schema v10; unchanged frozen/scientific catalogs and temporal implementation;
+real database logical SHA-256
+`26139cafeffbde3ec08f6ef23c5069e75afb9cd69ffffb40be5c099006fedc4d`; counts
+of 1 body observation, 6 continuous activities, 3 custom equipment, 23 zone
+sync rows, 32 exercise-zone rows, 23 exercises, 19 max results, 28 performed
+sets, 31 occurrences, and 3 sessions; `PRAGMA integrity_check = ok`; clean
+foreign-key check; and an empty index. Current closeout capture locations are
+`/tmp/trainlog-knowledge-review-resume/final.patch` and
+`/tmp/trainlog-knowledge-review-resume/final-manifest.json`. The completed capture
+includes all 19 modified tracked files and 37 untracked files, validates their
+hashes and whitespace, and confirms the real index remained untouched and empty.
+
+## Validation and retained evidence
+
+- 42 desktop tests; eight Python knowledge tests; four Python temporal tests.
+- Android tests and debug assembly: 56 tests, zero failures/errors, one skipped
+  `RealAndroidV9BodyZonesMigrationTest` because `TRAINLOG_ANDROID_V9_FIXTURE`
+  is unavailable. No knowledge/temporal test is skipped.
+- Catalog, JSON and import-contract validators; project skill validator.
+- Three public C17 headers with `-pedantic-errors` and strict warning flags.
+- Two affected C tests under ASan/UBSan, plus the temporal Python test; independent
+  twelve-form production-API probe also passes under ASan/UBSan.
+- `git diff --check` passes. Manual MTP/device/visual checks remain unclaimed.
+
+Current logs/results and reusable runner include
+`/tmp/trainlog-knowledge-review-resume/full-results.json`,
+`android-results.json`, `android-counts.json`, `sanitizer-results.json`, and
+`probe-normal-results.json`.
+Reusable sanitizer build: `/tmp/trainlog-knowledge-validation/build`.
+Implementation record: `/tmp/trainlog-temporal-implementation.md`.
+
+The older `/tmp/trainlog-knowledge-final.*` files are a previous checkpoint,
+not current complete evidence. Durable contract/evidence is in this review
+directory when `/tmp` is absent.
+
+Known explicit limits: C output timestamp capacity 40 characters with explicit
+failure for a selected longer value; no leap-event table (`:60` rejected);
+O(N) metadata scan per page; older unrelated lexical history readers remain
+future scope. No source-text normalization, migration or history rewrite is
+part of this correction.

docs/reviews/training_knowledge_v1_scientific_review.md

@@ -0,0 +1,154 @@
+# TRAINING KNOWLEDGE V1 — final scientific review
+
+Date: 2026-09-09. Result: **PASS within the documented scope and uncertainty**.
+No blocking scientific correction or confidence downgrade is required.
+
+This bounded review assessed the six canonical scientific catalogs against the
+previously researched source corpus, reviewed staging interpretations and
+`schema_supplement.json`. It also inspected the generated BODY ZONE audit,
+`docs/domain/` synthesis and Android's resolved-versus-conditional query guard.
+It is a scientific review, not a build, persistence or UI validation report.
+No production code, canonical catalog, database or persisted relation was edited
+by this review.
+
+## Coverage and fidelity
+
+| Canonical collection | Count | Finding |
+|---|---:|---|
+| Scientific references | 23 | Bibliography, identifiers, source types, limitations and claims preserved |
+| Functional muscle entities | 53 | Muscles, regions and groups remain distinct; membership overlap documented |
+| Joint actions | 35 | Actions, joint complexes, nominal planes and non-exhaustive contributors preserved |
+| Movement patterns | 26 | Authored conventions, French names, zones and anti-motion distinctions preserved |
+| Exercise identities | 23 | 17 resolved families, 4 conditional candidates, 2 unresolved warm-ups |
+| Equipment identities | 41 | 38 supplied and 3 custom; manufacturer and model remain null throughout |
+| BODY ZONE audit rows | 23 | 16 confirmed, 4 questionable, 3 unresolved; no proposed mutation |
+
+Scientific fields compare equal to the settled staging data plus the reviewed
+supplement after ignoring array order. References compare equal under the
+canonical field renaming. The rotary-torso compatibility representation follows
+`representation-decisions.md`: a symmetric UUID association marked
+`catalog_compatible_not_observed_occurrence` does not assert observed history.
+
+The exercise confidence distribution is 6 `high`, 11 `moderate` and 6
+`uncertain`. The equipment scientific-status distribution is 21
+`scientifically_documented`, 17 `mechanically_identified_anatomy_incomplete`
+and 3 `equipment_identity_uncertain`. Scientific documentation of a generic
+machine family does not certify its local model, trajectory or calibration.
+
+## All six high-confidence exercise mappings
+
+Here `high` applies to the explicitly limited movement-family interpretation:
+the relevant action, plausible muscle roles and coarse BODY ZONE projection.
+It does not assert measured local force shares, a known manufacturer, exact
+resistance curves, equal adaptation or transferable MAX. The source hierarchy
+supports these bounded claims without requiring an EMG study for every simple
+anatomical action.
+
+| Stable exercise ID | Identified variant and equipment | Actions, roles and zones | Evidence assessment |
+|---|---|---|---|
+| `ex_01ff06dd-ad00-46ee-9b46-ed32cabedbef` | Hip adduction; recorded `hip_adduction` context | Resisted hip adduction; adductor group; `thighs` | High retained. Established adductor function supports the family mapping; individual contributions depend on hip/knee position. |
+| `ex_1872246a-39ae-44dc-b58d-f87e90ca49ab` | Leg extension; recorded `leg_extension` context | Knee extension; quadriceps; `thighs` | High retained. Rectus femoris/vasti distinction and alignment/hip-angle limitations are explicit. |
+| `ex_617007f9-7420-4408-91b9-8ffb77900f13` | Seated Leg; recorded `seated_leg_curl` establishes curl interpretation beyond the ambiguous label | Knee flexion; hamstrings with possible gastrocnemius assistance; `thighs` | High retained. Equipment provenance identifies the variant; no name heuristic or identity merge. |
+| `ex_a1ef5047-b44b-4c64-a6ed-c7a3bc13b163` | Seated leg curl; recorded `seated_leg_curl` | Knee flexion; hamstrings with gastrocnemius assistance; `thighs` | High retained. Hip-flexion length effect is confined to biarticular hamstrings; no outcome equivalence claimed. |
+| `ex_d7398d9f-d928-4d2e-94e9-74e201da55c5` | Prone leg curl; recorded `prone_leg_curl` | Knee flexion; hamstrings with position-dependent gastrocnemius assistance; `thighs` | High retained. Prone/seated context remains distinct, without an unverified exact hip angle. |
+| `ex_ec619fc2-4685-4044-873c-86764bd4a0fe` | Arm curl; recorded `arm_curl` | Elbow flexion; biceps/brachialis primary, brachioradialis secondary, plausible forearm stabilization; `arms` | High retained for qualitative family roles. Grip/shoulder support limitations prevent a universal quantitative ranking. |
+
+The first five rows use established lower-limb anatomy. OpenStax's main text
+identifies the adductors, knee extensors and knee flexors and distinguishes
+lower-leg plantarflexors. Inconsistent image alternative text is not used to
+reverse an anatomical action.
+[OpenStax lower-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-6-appendicular-muscles-of-the-pelvic-girdle-and-lower-limbs)
+
+The seated/prone curl distinction also has an intervention source which
+explicitly considers biarticular hamstrings and biceps femoris short head.
+Its longitudinal finding is not generalized to all people, machines or
+strength tasks. [Maeo and colleagues](https://pubmed.ncbi.nlm.nih.gov/33009197/)
+
+The curl mapping is supported by the anatomical elbow-flexor and forearm
+functions. Its primary/secondary labels are qualitative exercise interpretation,
+not ratios established by the textbook.
+[OpenStax upper-limb anatomy](https://openstax.org/books/anatomy-and-physiology-2e/pages/11-5-muscles-of-the-pectoral-girdle-and-upper-limbs)
+
+## Representation checks
+
+- Custom Chest press, Converting chest press and Abdominal retain
+  `interpretation: null`. Their useful candidate interpretations remain
+  explicitly conditional and uncertain. Chest candidates include the reviewed
+  `pectoralis_major` / `horizontal_push` interpretation only under stated
+  execution assumptions.
+- Marche retains a conditional gait interpretation, without creating an
+  occurrence-to-treadmill association or assigning a persisted full-body zone.
+  Both generic warm-up identities have neither ordinary nor candidate mapping.
+- Rear Delt is linked to its actual UUID on the combined apparatus. Pec Fly
+  is a separate unlinked capability. Assisted dip and assisted chin/pull-up
+  are separate unlinked capabilities with empty actual exercise-ID lists.
+  No capability or legacy manifest slug becomes a phantom exercise identity.
+- The four added aggregates preserve member and overlap semantics. The
+  functional hip-flexor and spinal-stabilizer groups are non-exhaustive.
+  Their action lists describe member capabilities, not actions shared by every
+  member. Anatomical regions remain `muscle_region`.
+- Action contributors are non-exhaustive; missing a possible contributor does
+  not claim absence of participation. Nominal planes are not trajectory
+  constraints. Anti-extension, anti-rotation and lateral stabilization have
+  no invented dynamic joint action and remain authored task conventions.
+- Manufacturer/model fields are null for every physical identity. Example
+  manufacturer references remain mechanics/identification examples rather
+  than anatomical outcome evidence or local equipment identification.
+- EMG findings remain distinct from intervention evidence and do not establish
+  force shares, hypertrophy rankings or universal primary-muscle status.
+  [Vigotsky and colleagues](https://pubmed.ncbi.nlm.nih.gov/29354060/)
+- BODY ZONE audit `confirmed` means the existing coarse projection is
+  scientifically compatible under the stated interpretation. It does not
+  certify observed technique. `questionable` preserves conditional geometry
+  or identity; it is not a direction to mutate persistence.
+- Android's inspected ordinary query path obtains only
+  `resolved_family_variant_limited` interpretations. Conditional/null records
+  cannot enter ordinary muscle, pattern or scientific-zone matches through
+  that guard. Broader runtime behavior remains the engineering review's scope.
+
+## Remaining uncertainty and handoff
+
+Exact local manufacturer/model, resistance curves, range settings and detailed
+trajectories remain unresolved. Custom apparatus require actual execution
+confirmation. Back Extension requires pelvic-restraint and spinal-versus-hip
+motion evidence. Row/pulldown contributions depend on arm path and support;
+hip-abduction contributions depend on hip angle. Rotary torso compatibility
+must not be presented as a historical occurrence. Missing Pec Fly and assisted
+exercise identities require real catalog creation or identity evidence before
+linkage. Generic warm-ups require their actual activity sequences.
+
+The scientific foundation supports contextual MAX interpretation only.
+Assistance is not external resistance; no conversion, estimated 1RM, percentage
+prescription, universal recovery duration or exercise-equivalence guarantee is
+justified by these catalogs.
+
+These uncertainties require observation or future scoped research. They do
+not require xhigh escalation solely because the catalog is large. Scientific
+implementation handoff: preserve the reviewed distinctions and current
+confidence levels, and complete engineering validation separately. No BODY
+ZONE migration is recommended.
+
+## Reviewed catalog content hashes
+
+SHA-256 values identify the exact reviewed catalog bytes. A subsequent change
+to scientific fields requires a bounded delta review.
+
+| File | SHA-256 |
+|---|---|
+| `science-references-v1.json` | `ceef31455e0c1da1bc383d4b3a7dc63198cfb62ba0f80e26bc79bae77abb7836` |
+| `muscles-v1.json` | `21cb08c9c2ca393d5c87cf73a607175b7c267061687767912a0c92769be5ff4b` |
+| `joint-actions-v1.json` | `a66737f4419bcc4040d0abe5ca525b2a185e80f0bcf679091222eb540b3fd8a1` |
+| `movement-patterns-v1.json` | `ddd4e3522ebcb4df4618ed2a952fdb5834c42e21e7220a0afd69ab8f3437f072` |
+| `exercise-knowledge-v1.json` | `30ee4f400626af9a39c0547ae28360db94e76736462501e113f57721429ef43f` |
+| `equipment-knowledge-v1.json` | `b96ca23026fa1f0eb94bc8c0b2a586951130919db5342ee5d60ff1353725acb2` |
+
+### Subsequent representation verification
+
+The authored BODY ZONE audit was subsequently embedded in each exercise record
+so both platforms and the audit generator can read it from the canonical source.
+An automated comparison verified that removing only `body_zone_audit` reproduces
+the exact reviewed exercise-catalog SHA-256 above. All 23 embedded annotations
+also equal the reviewed audit data after the documented status and array-order
+normalization. No scientific assertion or confidence changed. The complete
+exercise catalog now has SHA-256
+`a2c72e4743dc5dc24d1a7bbf9e8234442adc2d5ef6882f15cadea1cb9629d647`.

docs/reviews/training_knowledge_v1_temporal_contract.md

@@ -0,0 +1,139 @@
+# Training Knowledge V1 — temporal contract and correction
+
+Date: 2026-09-09. Contract analysis: advisor / Terra-high, completed.
+Temporal implementation and executable validation: PASS. Independent temporal
+delta review: `TEMPORAL_DELTA_REVIEW=PASS`, with no findings or repairs. The
+initial full-tranche audit found non-temporal defects and its authorized repair
+chain completed. Independent bounded verification and fresh final validation
+subsequently passed; `TRAINING_KNOWLEDGE_V1=PASS`. This record preserves the
+temporal contract and does not expand scientific scope or declare a frozen
+format decision.
+
+## Authority and compatibility
+
+The frozen [exchange timestamp rule](../exchange_format.md#7-session-timestamps)
+requires RFC 3339 / ISO 8601 date-time text with an explicit UTC offset. The V1
+schema declares `format: date-time`. The operational reader profile is:
+
+```text
+YYYY-MM-DD[Tt]HH:MM[:SS[.digits]](Z|z|±HH:MM)
+```
+
+It uses extended Gregorian dates in years 0001–9999, valid calendar days,
+hours 00–23, minutes 00–59 and ordinary seconds 00–59. Fractions require
+seconds and at least one ASCII digit after a dot. Every fractional digit
+participates in comparison; trailing zeros do not change an instant.
+
+Numeric offsets range through ±23:59. Lowercase `t`/`z` and the offset grammar
+follow [RFC 3339 §5.6](https://www.rfc-editor.org/rfc/rfc3339#section-5.6).
+`-00:00` compares as a zero UTC offset while its distinct source spelling is
+preserved. Omitted seconds are an application compatibility requirement:
+Android persists `OffsetDateTime.now().toString()`, which can omit seconds
+when seconds and fractional seconds are both zero.
+
+Basic dates, ISO week dates, spaces instead of `T/t`, comma fractions, compact
+offsets and hour-only offsets are **NOT CONTRACTUALLY REQUIRED**. They are
+rejected by the selected profile. The observed acceptance of these forms by
+Python `datetime.fromisoformat()` is not normative evidence: this environment's
+jsonschema 4.26.0 has no registered default `date-time` checker. No Trainlog
+writer, pre-existing fixture or inspected real session establishes a need for
+those extensions. All three real session timestamps match the selected
+profile and fit the existing C result fields.
+
+Years outside 0001–9999 and `:60` remain outside the established operational
+admission. Supporting real leap seconds would require a verified shared
+leap-event schedule and matching validator behavior; the repair does not
+guess a conversion or collapse a leap second into the following minute.
+[RFC 3339 §5.7](https://www.rfc-editor.org/rfc/rfc3339#section-5.7)
+
+## Root cause and implementation
+
+The old cursor guard, SQLite `julianday()`, Java `OffsetDateTime.parse()` and
+Python's ISO parser had different languages and precision. A `+14:30` source
+produced a rejected cursor. SQLite returned NULL for valid `+15:00` and
+lowercase-`t` values, allowing records to disappear. Floating-point Julian
+days also did not preserve arbitrary fractional precision.
+
+The C and Kotlin readers now parse source text into an integer UTC second
+and an exact fractional digit sequence. Both occurrence pagination and the
+new latest-explicit-MAX reader order descending by:
+
+```text
+(represented UTC instant, session_id bytes, entry_id bytes)
+```
+
+The cursor applies the same comparison exclusively. Equal instants with
+different offsets or fractional trailing zeros use identity tie breakers,
+never timestamp-text tie breakers. Cursor emission preserves the source
+timestamp and both IDs. C supports aliasing the input and output cursor.
+
+Each read scans all matching metadata candidates and retains at most
+`limit + 1` occurrence candidates, or one MAX candidate. Only selected rows
+are hydrated. It does not use a timestamp SQL predicate, SQL temporal order
+or SQL limit before comparison. Scan work is O(N) for the exercise history;
+retained candidate count is O(limit), with no global dataset-size cap. Each
+read has a snapshot covering selection and hydration. Separate page calls
+retain the existing current-data semantics, without a cross-call snapshot.
+
+Malformed caller cursors fail before the scan. Malformed matching stored
+timestamps cause an explicit database/consistency error; they are not
+silently skipped. Source timestamps are never rewritten or normalized in
+persistent storage.
+
+The C public timestamp fields retain their existing 40-character capacity.
+Longer valid SQLite text is compared with full precision; if selected for
+output, it produces `DATABASE_ERROR`. It is never truncated or silently
+omitted. Android can return longer strings. This is an existing C API
+representation limit, not a new wire precision bound. Arbitrary-length C
+output would require a separate ownership/length API change.
+
+The Python validator uses the same grammar and exact chronology. It overrides
+only `date-time` on a fresh per-validation format-checker instance, preserving
+other checks and caller/global state. The document path therefore admits
+Android's no-seconds form even when an optional strict RFC checker is present.
+
+## Regression and validation evidence
+
+Production C and Android tests cover mixed representations, page size one,
+emitted cursors, continuation, exhaustion, no duplicates, chronological order,
+equal-instant session/entry ties, near-equal fractions, malformed cursors and
+stored values, and latest-MAX selection. C also checks selected and unselected
+timestamps beyond its fixed output capacity. Python tests cover grammar,
+exact chronology, date/offset bounds and missing/strict/permissive optional
+format checkers through actual document validation.
+
+An independent C production-API probe exercises twelve admitted spellings:
+`+02:00`, `+14:30`, `+15:00`, lowercase `t`, `Z`, lowercase `t/z`, omitted
+seconds, `+23:59`, `-23:59`, `-00:00`, and two long fractional forms. Each case
+uses two sessions in a temporary database, page size one, exact source/cursor
+equality checks and traversal through exhaustion. It passes normally and
+under ASan/UBSan.
+
+Full validation passes: 42 desktop tests; eight knowledge Python tests; four
+timestamp Python tests; JSON/import/catalog validators; three standalone C17
+headers; Android unit tests and debug assembly (56 tests, zero failures or
+errors, one pre-existing missing-fixture skip); two targeted C sanitizer tests
+plus the timestamp Python test; project skill validation; `git diff --check`.
+Logs and reusable runner: `/tmp/trainlog-temporal-validation/`.
+
+The independent temporal reviewer covered grammar, calendar validity, offset
+ranges, exact fractions, bytewise ties, cursor aliasing and exclusivity, source
+capacity, snapshot behavior, and Android/Python parity. It ran the targeted
+Meson `training_context` and `timestamp_validation` tests and all four Python
+temporal tests. It reported no temporal defect, repair, or unresolved temporal
+boundary.
+
+## Preserved boundaries and remaining scope
+
+Desktop schema v11, Android schema v10, stable IDs, real database logical
+contents and counts, BODY ZONES, equipment and all scientific catalog bytes
+remain unchanged. Scientific review stays valid without a scientific delta.
+No staging, commit, push, database deletion, migration, device installation,
+uninstall, keystore change or history rewrite occurred.
+
+Older general history/export/MAX-list readers with lexical timestamp ordering
+are outside this bounded new-reader repair. Leap-event support, arbitrary-
+length C output and manual device/MTP/visual checks remain explicit future or
+manual work. The previously recorded secondary-zone-ID TUI presentation
+inconsistency remains nonblocking. Final-review repair verification and fresh
+final validation completed with `TRAINING_KNOWLEDGE_V1=PASS`.

docs/reviews/training_knowledge_v1_temporal_review.md

@@ -0,0 +1,27 @@
+# Training Knowledge V1 — independent temporal delta review
+
+Date: 2026-09-09. Result: `TEMPORAL_DELTA_REVIEW=PASS`.
+
+This isolated review found no temporal defect and required no repair. It does
+not declare the wider `TRAINING_KNOWLEDGE_V1` tranche PASS or FROZEN.
+
+## Scope and evidence
+
+The reviewer independently examined the settled temporal reader contract for:
+
+- accepted grammar, calendar validation, and numeric offset range;
+- exact fractional-second comparison and bytewise stable-ID ties;
+- exclusive cursor behavior and C input/output cursor aliasing;
+- source-text output capacity and explicit failure behavior;
+- selection/hydration snapshot boundaries; and
+- parity between C, Android, and Python document validation.
+
+It ran Meson `training_context` and `timestamp_validation` tests and all four
+Python temporal tests. No temporal semantic discrepancy, repair, or remaining
+temporal finding was reported.
+
+The settled admitted source profile, ordering, cursor, snapshot, and explicit
+error semantics remain verbatim in the [temporal contract](training_knowledge_v1_temporal_contract.md).
+The separate final-review repair verification and final validation subsequently
+passed. `TRAINING_KNOWLEDGE_V1=PASS`; this record remains limited to the
+independent temporal review and does not expand scientific scope.

docs/roadmap.md

@@ -31,7 +31,7 @@ BODY_ZONES_V1=PASS
 BODY_ZONE_SYNC_V1=PASS
 BODY_ZONES_ANDROID_DEVICE_VALIDATION=PASS
 
-DESKTOP_TESTS=39/39 PASS
+DESKTOP_TESTS=42/42 PASS (recorded validation checkpoint)
 TUI_NOTCURSES_V1=PASS
 NCURSESW_REMOVED_FROM_ACTIVE_TUI=PASS
 NOTCURSES_TRUECOLOR_THEME=PASS
@@ -63,6 +63,18 @@ descendant-aware filters, unclassified history and one explicit-conflict sync
 companion. It does not implement a session generator, custom zones or proposed
 loads.
 
+`TRAINING_KNOWLEDGE_V1=PASS` is read-only infrastructure. Scientific review,
+independent temporal review, final engineering review, repair verification, and
+final executable validation passed. The temporal contract preserves source text
+and exact C/Android chronological pagination; one bounded repair chain closed
+the audit's stale-documentation, Android-loader, Meson-input, and role-only C
+query findings.
+It provides catalog-backed scientific lookup and
+runtime context composition without prescriptions, schema changes or catalog
+seeding. The documented future session/program input pipeline is deliberately
+not a roadmap gate or an implemented generator; see
+[Training knowledge system V1](domain/knowledge_system.md).
+
 `EXERCISE_EDIT_V1` is a completed capture correction: Android permits
 stable-ID renames, protects referenced profiles, and reconciles same-ID display
 metadata without duplicates. `ANDROID_BANNER_PARITY_V1` is a presentation-only

docs/tests.md

@@ -528,3 +528,48 @@ inspection/pull. Applying the validated state to both real stores and running
 the two transport directions remains an out-of-sandbox hardware validation.
 
 Detailed retained evidence: [Android draft execution record](reviews/android_session_draft_v1_resume.md).
+
+## 15. Training knowledge V1 validation checkpoint
+
+The current implementation passes the training-knowledge catalog
+validator, eight Python generation/catalog regressions, 42 desktop Meson tests,
+and standalone C17 public-header checks for `training_knowledge.h`,
+`training_context.h` and `database.h`. A targeted AddressSanitizer/Undefined-
+Behavior-Sanitizer build passed the two new desktop API tests. Android
+`testDebugUnitTest assembleDebug` passed with 56 tests, zero failures/errors,
+and one skipped real-v9 fixture test because `TRAINLOG_ANDROID_V9_FIXTURE` was
+not available. The TUI knowledge screen's UTF-8 cell-aware scrolling was tested
+at the 72x20 minimum terminal.
+
+The catalog validator, JSON validator and import-contract validator also pass.
+`git diff --check` passes. Scientific-review metadata and
+catalog hashes were verified separately.
+
+Temporal regressions now cover the original `+14:30`, `+15:00` and lowercase-`t`
+failure, lowercase `z`, omitted seconds, high/negative offsets, exact fractions,
+stable identity ties, cursor reuse, traversal through exhaustion and malformed
+stored values. C tests also cover selected/unselected values beyond its fixed
+output capacity. Four Python tests cover the explicit grammar and exact
+chronology, including actual document admission with missing, strict and
+permissive optional JSON Schema format checkers. An independent twelve-form
+production C probe passes normally and under ASan/UBSan.
+
+Fresh final validation passed: strict build; 42 Meson tests; eight Python
+knowledge tests; four Python temporal tests; knowledge, JSON, and import
+validators; three strict C17 headers; affected C knowledge/context tests under
+ASan/UBSan plus Python timestamp validation; normal and sanitized independent
+12-form temporal probes; Android 56 tests with zero failures/errors and one
+known missing real-v9 fixture skip; Java 17 debug assembly; skill validation;
+and `git diff --check`. Generated C is byte-identical with SHA-256
+`e8c099f67eb111d61621b5d76592c049823af5508d43e73ec646f22e4c377fca`; all six
+Android assets are byte-identical.
+
+`TRAINING_KNOWLEDGE_V1=PASS`. The independent temporal review passed with no
+findings; its coverage included parser grammar and limits, exact chronology,
+ties, cursor semantics, source capacity, snapshots, and Android/Python parity.
+The initial full audit's stale temporal documentation, Android loader, Meson
+dependency, and C role-only query findings were repaired and independently
+verified. See the [temporal contract](reviews/training_knowledge_v1_temporal_contract.md)
+for the established contract. No real Android install,
+manual TUI visual exercise, or manual MTP hardware validation was performed
+for Training Knowledge V1.

docs/tui.md

@@ -465,3 +465,18 @@ Proportions can display:
 
 All estimates are explicitly labeled as estimates. No result is converted into
 a medical or diagnostic classification.
+
+## 18. Training knowledge infrastructure
+
+The desktop core includes immutable `training_knowledge.h` catalog access and
+read-only `training_context.h` composition. The context uses real persisted IDs
+and returns persisted zones, optional science, compatible equipment, latest
+explicit MAX and bounded chronological history; it does not write data. The
+TUI provides a UTF-8 cell-aware scrolling knowledge screen, tested at the 72x20
+minimum terminal. V1 provides no prescription, generator or scoring flow. Its
+lifecycle is `TRAINING_KNOWLEDGE_V1=PASS`. Its occurrence and latest-MAX readers use the settled
+temporal contract: exact instant/fraction ordering, bytewise stable-ID ties,
+exclusive source-text cursors, bounded selection before hydration, and one
+read snapshot per call. Malformed cursors and matching stored timestamps fail
+explicitly. The full-tranche audit repair chain and independent verification
+passed. See [Training knowledge system V1](domain/knowledge_system.md).

tests/fixtures/invalid/timestamp-space-separator.json

@@ -0,0 +1,10 @@
+{
+  "format": "trainlog",
+  "version": 1,
+  "exercises": [],
+  "session": {
+    "session_id": "invalid-temporal-space",
+    "started_at": "2026-09-05 18:34:12+02:00",
+    "exercises": []
+  }
+}

tests/fixtures/valid/temporal-extended-forms.json

@@ -0,0 +1,11 @@
+{
+  "format": "trainlog",
+  "version": 1,
+  "exercises": [],
+  "session": {
+    "session_id": "temporal-extended-forms",
+    "started_at": "2026-09-05t18:34:12.12345678901234567890z",
+    "ended_at": "2026-09-06T18:34+23:59",
+    "exercises": []
+  }
+}

tests/test_timestamp_validation.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+"""Focused frozen timestamp grammar and exact chronology regressions."""
+import sys
+import unittest
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "tools"))
+from validate_json import TrainlogSemanticError, parse_timestamp  # noqa: E402
+import validate_json as validation  # noqa: E402
+
+class TimestampValidationTest(unittest.TestCase):
+    def test_document_admission_does_not_depend_on_optional_format_checker(self):
+        schema = validation.load_json(validation.SCHEMA_PATH)
+        # Simulate missing, stricter RFC-only, and permissive optional checkers.
+        # Exercise the actual document path, including schema checks before semantics.
+        for mode in ("missing", "strict", "permissive"):
+            checker = validation.jsonschema.FormatChecker()
+            checker.checkers.pop("date-time", None)
+            if mode != "missing":
+                checker.checks("date-time")(
+                    lambda value, strict=mode == "strict": not strict
+                    or not isinstance(value, str)
+                    or len(value) > 19 and value[16] == ":"
+                )
+            original_checks = checker.checkers.copy()
+            validator = validation.jsonschema.Draft202012Validator(schema, format_checker=checker)
+            with self.subTest(mode=mode):
+                self.assertEqual([], validation.validate_document(
+                    validator, validation.VALID_FIXTURE_DIR / "temporal-extended-forms.json"))
+                self.assertTrue(validation.validate_document(
+                    validator, validation.INVALID_FIXTURE_DIR / "timestamp-space-separator.json"))
+                self.assertEqual(original_checks, checker.checkers)
+
+    def test_selected_grammar(self):
+        for value in ("0001-01-01T00:00Z", "2000-02-29t23:59:59.00000000000000000001z",
+                      "2026-09-05T18:34+23:59", "2026-09-05T18:34:12-00:00",
+                      "9999-12-31T23:59:59.9-23:59"):
+            with self.subTest(value=value): self.assertIsNotNone(parse_timestamp(value, "probe"))
+
+    def test_platform_only_forms_rejected(self):
+        for value in ("0000-01-01T00:00:00Z", "2026-02-29T00:00:00Z",
+                      "2026-09-05 18:34:12+02:00", "20260905T183412+0200",
+                      "2026-W36-5T18:34:12+02:00", "2026-09-05T18:34:12,5+02:00",
+                      "2026-09-05T18:34:12+0200", "2026-09-05T18:34:12+02",
+                      "2026-09-05T18:34:12.5", "2026-09-05T18:34.5Z",
+                      "2026-09-05T18:34:60Z", "2026-09-05T18:34:12+24:00"):
+            with self.subTest(value=value), self.assertRaises(TrainlogSemanticError):
+                parse_timestamp(value, "probe")
+
+    def test_exact_fraction_and_rollover(self):
+        earlier = parse_timestamp("2026-01-01T00:00:00.12345678901234567890Z", "probe")
+        later = parse_timestamp("2026-01-01T00:00:00.12345678901234567891Z", "probe")
+        equal = parse_timestamp("2026-01-01T00:00:00.1234567890123456789000Z", "probe")
+        self.assertLess(earlier, later); self.assertEqual(earlier, equal)
+        self.assertLess(parse_timestamp("2025-12-31T09:14:59Z", "probe"),
+                        parse_timestamp("2026-01-01T00:15:00+15:00", "probe"))
+
+if __name__ == "__main__": unittest.main()

tests/test_training_knowledge.py

@@ -0,0 +1,133 @@
+import copy
+import importlib.util
+import json
+import subprocess
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+SPEC = importlib.util.spec_from_file_location(
+    "validate_training_knowledge", ROOT / "tools" / "validate_training_knowledge.py"
+)
+MODULE = importlib.util.module_from_spec(SPEC)
+assert SPEC.loader is not None
+SPEC.loader.exec_module(MODULE)
+
+
+class TrainingKnowledgeValidationTest(unittest.TestCase):
+    files = (
+        "body-zones-v1.json", "equipment-v1.json", "science-references-v1.json", "muscles-v1.json",
+        "joint-actions-v1.json", "movement-patterns-v1.json",
+        "exercise-knowledge-v1.json", "equipment-knowledge-v1.json",
+        "training-knowledge-audit-v1.json",
+    )
+
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        self.catalog = Path(self.temp.name)
+        for filename in self.files:
+            (self.catalog / filename).write_bytes((ROOT / "catalog" / filename).read_bytes())
+
+    def tearDown(self):
+        self.temp.cleanup()
+
+    def mutate(self, filename, operation):
+        path = self.catalog / filename
+        value = json.loads(path.read_text(encoding="utf-8"))
+        operation(value)
+        path.write_text(json.dumps(value, ensure_ascii=False), encoding="utf-8")
+
+    def assert_invalid(self):
+        with self.assertRaises(MODULE.ValidationError):
+            MODULE.validate(self.catalog)
+
+    def test_valid_and_deterministic_generation(self):
+        MODULE.validate(self.catalog)
+        first = self.catalog / "first.c"
+        second = self.catalog / "second.c"
+        command = [sys.executable, str(ROOT / "tools" / "generate_training_knowledge.py"),
+                   str(self.catalog)]
+        subprocess.run(command + [str(first)], check=True)
+        subprocess.run(command + [str(second)], check=True)
+        self.assertEqual(first.read_bytes(), second.read_bytes())
+        audit = self.catalog / "regenerated-audit.json"
+        subprocess.run([sys.executable, str(ROOT / "tools" / "generate_training_knowledge_audit.py"),
+                        str(self.catalog), str(audit)], check=True)
+        self.assertEqual(audit.read_bytes(), (self.catalog / "training-knowledge-audit-v1.json").read_bytes())
+        markdown = self.catalog / "audit.md"
+        subprocess.run([sys.executable, str(ROOT / "tools" / "generate_training_knowledge_audit.py"),
+                        str(self.catalog), str(markdown), "--markdown"], check=True)
+        self.assertEqual(markdown.read_bytes(), (ROOT / "docs/domain/knowledge_audit.md").read_bytes())
+
+    def test_unknown_nested_keys_and_types_are_validation_errors(self):
+        mutations = [
+            lambda root: root["equipment"][0].__setitem__("unexpected_review_key", True),
+            lambda root: root["equipment"][0]["capabilities"][0].__setitem__("requirements", None),
+            lambda root: root["equipment"][0].__setitem__("requires_actual_exercise", 1),
+        ]
+        original = (self.catalog / "equipment-knowledge-v1.json").read_bytes()
+        for mutation in mutations:
+            (self.catalog / "equipment-knowledge-v1.json").write_bytes(original)
+            self.mutate("equipment-knowledge-v1.json", mutation)
+            self.assert_invalid()
+
+    def test_additive_non_runtime_reference_is_valid(self):
+        def add(root):
+            row = copy.deepcopy(root["references"][-1])
+            row["ref_id"] = "zz_additive_provenance"
+            row["title"] = "Additional provenance record"
+            root["references"].append(row)
+        self.mutate("science-references-v1.json", add)
+        MODULE.validate(self.catalog)
+
+    def test_stale_generated_audit_is_rejected(self):
+        self.mutate("training-knowledge-audit-v1.json",
+                    lambda root: root["rows"][0].__setitem__("rationale", "stale generated value"))
+        self.assert_invalid()
+
+    def test_duplicate_object_key_rejected(self):
+        path = self.catalog / "science-references-v1.json"
+        text = path.read_text(encoding="utf-8")
+        path.write_text(text.replace('"version": 1', '"version": 1, "version": 1', 1), encoding="utf-8")
+        self.assert_invalid()
+
+    def test_version_order_enum_role_and_cross_reference_mutations(self):
+        cases = [
+            ("muscles-v1.json", lambda root: root.__setitem__("version", 2)),
+            ("muscles-v1.json", lambda root: root["muscles"].reverse()),
+            ("muscles-v1.json", lambda root: root["muscles"][0]["joint_action_ids"].reverse()),
+            ("muscles-v1.json", lambda root: root["muscles"][0].__setitem__("confidence", "limited")),
+            ("exercise-knowledge-v1.json", lambda root: root["exercises"][0]["interpretation"].
+             __setitem__("primary_muscle_ids", ["missing_muscle"])),
+            ("exercise-knowledge-v1.json", lambda root: root["exercises"][0]["interpretation"].
+             __setitem__("secondary_muscle_ids", root["exercises"][0]["interpretation"]["primary_muscle_ids"])),
+            ("equipment-knowledge-v1.json", lambda root: root["equipment"][0]["capabilities"][0].
+             __setitem__("exercise_ids", ["ex_00000000-0000-4000-8000-000000000000"])),
+        ]
+        originals = {filename: (self.catalog / filename).read_bytes() for filename in self.files}
+        for filename, mutation in cases:
+            for restore, contents in originals.items():
+                (self.catalog / restore).write_bytes(contents)
+            self.mutate(filename, mutation)
+            self.assert_invalid()
+
+    def test_conditional_candidate_cannot_become_resolved_data(self):
+        def leak(root):
+            row = next(item for item in root["exercises"] if item["resolution_status"] == "conditional")
+            row["interpretation"] = copy.deepcopy(row["conditional_interpretation"])
+        self.mutate("exercise-knowledge-v1.json", leak)
+        self.assert_invalid()
+
+    def test_manufacturer_only_high_mapping_rejected(self):
+        def mutation(root):
+            root["equipment"][0]["confidence"] = "high"
+            root["equipment"][0]["evidence_type"] = "manufacturer_statement"
+        self.mutate("equipment-knowledge-v1.json", mutation)
+        self.assert_invalid()
+
+
+if __name__ == "__main__":
+    unittest.main()

tools/generate_training_knowledge.py

@@ -0,0 +1,223 @@
+#!/usr/bin/env python3
+"""Generate deterministic C tables and read/query functions from knowledge V1."""
+
+from __future__ import annotations
+
+import json
+import sys
+from pathlib import Path
+
+from validate_training_knowledge import ValidationError, validate
+
+
+def c(value) -> str:
+    if value is None:
+        return "NULL"
+    if isinstance(value, list):
+        value = "\n".join(value)
+    return json.dumps(value, ensure_ascii=False)
+
+
+def interp(value) -> str:
+    if value is None:
+        return "{NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL,NULL}"
+    keys = ("family_description", "action_ids", "pattern_ids", "primary_muscle_ids",
+            "secondary_muscle_ids", "stabilizer_muscle_ids", "primary_zone_id",
+            "secondary_zone_ids", "confidence", "evidence_type", "source_refs",
+            "variant_notes", "role_notes", "required_confirmation")
+    return "{" + ",".join(c(value.get(key)) for key in keys) + "}"
+
+
+def load(path: Path, key: str):
+    return json.loads(path.read_text(encoding="utf-8"))[key]
+
+
+def main() -> None:
+    if len(sys.argv) != 3:
+        raise SystemExit("usage: generate_training_knowledge.py CATALOG_DIR OUTPUT_C")
+    root, output = Path(sys.argv[1]), Path(sys.argv[2])
+    try:
+        validate(root)
+    except ValidationError as error:
+        raise SystemExit(str(error)) from error
+    refs = load(root / "science-references-v1.json", "references")
+    muscles = load(root / "muscles-v1.json", "muscles")
+    actions = load(root / "joint-actions-v1.json", "joint_actions")
+    patterns = load(root / "movement-patterns-v1.json", "movement_patterns")
+    exercises = load(root / "exercise-knowledge-v1.json", "exercises")
+    equipment = load(root / "equipment-knowledge-v1.json", "equipment")
+    lines = ['#include "trainlog/training_knowledge.h"',
+             '#include "trainlog/body_zone_catalog.h"', "#include <string.h>", ""]
+    lines.append("static const TrainlogKnowledgeReference references[] = {")
+    for row in refs:
+        lines.append("  {%s}," % ",".join((c(row["ref_id"]), c(row["title"]),
+            c(row["authors_or_organization"]), str(row["year"] or 0), c(row["type"]),
+            c(row["url"]), c(row["doi"]), c(row["pmid"]), c(row["topics"]),
+            c(row["notes"]), c(row["limitations"]), c(row["accessed_on"]),
+            c(row.get("publication_note")))))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeBodyZoneAudit audits[] = {")
+    for row in exercises:
+        audit = row["body_zone_audit"]
+        interpretation = row["interpretation"] or row["conditional_interpretation"]
+        fields = (row["exercise_id"], audit["status"], audit["severity"], audit["rationale"],
+                  audit["confidence"], audit["source_refs"], audit["existing_primary_zone_id"],
+                  audit["existing_secondary_zone_ids"],
+                  None if interpretation is None else interpretation["primary_zone_id"],
+                  [] if interpretation is None else interpretation["secondary_zone_ids"],
+                  audit["proposed_mutation"])
+        lines.append("  {%s}," % ",".join(c(value) for value in fields))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeMuscle muscles[] = {")
+    for row in muscles:
+        lines.append("  {%s}," % ",".join(c(row.get(key)) for key in (
+            "muscle_id", "display_name", "display_name_fr", "entity_type", "anatomical_group",
+            "aggregate_group_id", "member_muscle_ids", "body_zone_ids", "joint_action_ids",
+            "primary_actions", "primary_actions_semantics", "functional_notes", "confidence", "evidence_type", "source_refs",
+            "overlap_warning")))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeJointAction actions[] = {")
+    for row in actions:
+        lines.append("  {%s}," % ",".join(c(row.get(key)) for key in (
+            "action_id", "display_name_fr", "definition", "anatomical_region", "joint_complex",
+            "principal_plane", "plane_notes", "contributing_muscle_ids", "contributor_semantics", "confidence",
+            "evidence_type", "source_refs", "notes")))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeMovementPattern patterns[] = {")
+    for row in patterns:
+        lines.append("  {%s}," % ",".join(c(row.get(key)) for key in (
+            "pattern_id", "display_name_fr", "definition", "parent_pattern_id", "typical_action_ids",
+            "typical_body_zone_ids", "body_zone_semantics", "confidence", "evidence_type", "source_refs", "notes")))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeInterpretation exercise_interpretations[] = {")
+    for row in exercises:
+        lines.append("  %s," % interp(row["interpretation"]))
+    lines.append("};")
+    lines.append("static const TrainlogKnowledgeInterpretation exercise_conditionals[] = {")
+    for row in exercises:
+        lines.append("  %s," % interp(row["conditional_interpretation"]))
+    lines.append("};")
+    lines.append("static const TrainlogExerciseKnowledge exercises[] = {")
+    for index, row in enumerate(exercises):
+        resolved = "&exercise_interpretations[%d]" % index if row["interpretation"] else "NULL"
+        conditional = "&exercise_conditionals[%d]" % index if row["conditional_interpretation"] else "NULL"
+        fields = [c(row.get(key)) for key in ("exercise_id", "exercise_name", "resolution_status",
+            "confidence", "equipment_ids", "identity_evidence", "identity_evidence_type", "source_refs",
+            "limitations", "equipment_link_status")]
+        lines.append("  {%s,%s,%s}," % (",".join(fields), resolved, conditional))
+    lines.append("};")
+    lines.append("static const TrainlogEquipmentKnowledge equipment[] = {")
+    for row in equipment:
+        fields = [c(row.get(key)) for key in ("equipment_id", "manufacturer", "model", "identification_status",
+            "scientific_status", "scientific_status_scope", "catalog_type", "catalog_load_semantics", "mechanics", "confidence",
+            "evidence_type", "source_refs", "limitations", "audit_note")]
+        fields.append("true" if row["requires_actual_exercise"] else "false")
+        lines.append("  {%s}," % ",".join(fields))
+    lines.append("};")
+    capabilities = [(row["equipment_id"], cap) for row in equipment for cap in row["capabilities"]]
+    lines.append("static const TrainlogKnowledgeInterpretation capability_interpretations[] = {")
+    for _, cap in capabilities:
+        lines.append("  %s," % interp(cap["interpretation"]))
+    lines.append("};")
+    lines.append("static const TrainlogEquipmentCapability capabilities[] = {")
+    for index, (equipment_id, cap) in enumerate(capabilities):
+        pointer = "&capability_interpretations[%d]" % index if cap["interpretation"] else "NULL"
+        lines.append("  {%s,%s,%s,%s,%s,%s}," % (c(equipment_id), c(cap["display_name"]),
+            c(cap["exercise_ids"]), c(cap["link_status"]), c(cap["requirements"]), pointer))
+    lines.append("};")
+    lines.append(r'''
+static bool list_has(const char *list, const char *id) {
+  size_t length;
+  const char *at;
+  if (list == NULL || id == NULL || id[0] == '\0') return false;
+  length = strlen(id);
+  at = list;
+  while (*at != '\0') {
+    const char *end = strchr(at, '\n');
+    size_t item_length = end == NULL ? strlen(at) : (size_t)(end - at);
+    if (item_length == length && memcmp(at, id, length) == 0) return true;
+    if (end == NULL) break;
+    at = end + 1;
+  }
+  return false;
+}
+#define DEFINE_ACCESSORS(prefix,type,array,key) \
+size_t prefix##_count(void) { return sizeof(array) / sizeof(array[0]); } \
+const type *prefix##_at(size_t index) { return index < prefix##_count() ? &array[index] : NULL; } \
+const type *prefix##_lookup(const char *id) { size_t i; if (id == NULL || id[0] == '\0') return NULL; \
+  for (i=0; i<prefix##_count(); ++i) { if (strcmp(array[i].key,id)==0) return &array[i]; } \
+  return NULL; }
+DEFINE_ACCESSORS(trainlog_knowledge_reference, TrainlogKnowledgeReference, references, ref_id)
+DEFINE_ACCESSORS(trainlog_knowledge_muscle, TrainlogKnowledgeMuscle, muscles, muscle_id)
+DEFINE_ACCESSORS(trainlog_knowledge_joint_action, TrainlogKnowledgeJointAction, actions, action_id)
+DEFINE_ACCESSORS(trainlog_knowledge_movement_pattern, TrainlogKnowledgeMovementPattern, patterns, pattern_id)
+DEFINE_ACCESSORS(trainlog_exercise_knowledge, TrainlogExerciseKnowledge, exercises, exercise_id)
+DEFINE_ACCESSORS(trainlog_knowledge_body_zone_audit, TrainlogKnowledgeBodyZoneAudit, audits, exercise_id)
+DEFINE_ACCESSORS(trainlog_equipment_knowledge, TrainlogEquipmentKnowledge, equipment, equipment_id)
+size_t trainlog_equipment_capability_count(void) { return sizeof(capabilities)/sizeof(capabilities[0]); }
+const TrainlogEquipmentCapability *trainlog_equipment_capability_at(size_t index) {
+  return index < trainlog_equipment_capability_count() ? &capabilities[index] : NULL;
+}
+const TrainlogKnowledgeInterpretation *trainlog_exercise_knowledge_conditional(const char *exercise_id) {
+  const TrainlogExerciseKnowledge *record = trainlog_exercise_knowledge_lookup(exercise_id);
+  return record == NULL ? NULL : record->conditional_interpretation;
+}
+static bool zone_matches(const TrainlogKnowledgeInterpretation *value, const char *zone, bool descendants) {
+  if (strcmp(value->primary_zone_id, zone) == 0 || list_has(value->secondary_zone_ids, zone)) return true;
+  if (!descendants) return false;
+  if (trainlog_body_zone_catalog_is_descendant(value->primary_zone_id, zone)) return true;
+  { const char *at = value->secondary_zone_ids;
+    while (at != NULL && *at != '\0') { const char *end = strchr(at, '\n'); char id[64];
+      size_t n = end == NULL ? strlen(at) : (size_t)(end-at); if (n >= sizeof(id)) return false;
+      memcpy(id,at,n); id[n]='\0'; if (trainlog_body_zone_catalog_is_descendant(id,zone)) return true;
+      if (end == NULL) break;
+      at=end+1; }
+  }
+  return false;
+}
+TrainlogStatus trainlog_exercise_knowledge_query(const TrainlogKnowledgeQuery *query,
+    const TrainlogExerciseKnowledge **output, size_t capacity, size_t *output_count) {
+  size_t i, count=0; const TrainlogKnowledgeQuery empty={0};
+  if (output_count == NULL || (capacity > 0 && output == NULL)) return TRAINLOG_STATUS_INVALID_ARGUMENT;
+  *output_count=0; if (query == NULL) query=&empty;
+  if (query->muscle_role < TRAINLOG_KNOWLEDGE_ROLE_ANY || query->muscle_role > TRAINLOG_KNOWLEDGE_ROLE_STABILIZER)
+    return TRAINLOG_STATUS_INVALID_ARGUMENT;
+  /* INVARIANT: a specific muscle role is meaningful only with a muscle ID.
+   * Keep the C query contract aligned with Android's paired filters. */
+  if (query->muscle_id == NULL && query->muscle_role != TRAINLOG_KNOWLEDGE_ROLE_ANY)
+    return TRAINLOG_STATUS_INVALID_ARGUMENT;
+  if (query->scientific_zone_id != NULL && trainlog_body_zone_catalog_lookup(query->scientific_zone_id) == NULL)
+    return TRAINLOG_STATUS_NOT_FOUND;
+  if (query->movement_pattern_id != NULL && trainlog_knowledge_movement_pattern_lookup(query->movement_pattern_id) == NULL)
+    return TRAINLOG_STATUS_NOT_FOUND;
+  if (query->muscle_id != NULL && trainlog_knowledge_muscle_lookup(query->muscle_id) == NULL)
+    return TRAINLOG_STATUS_NOT_FOUND;
+  if (query->available_equipment_id != NULL && trainlog_equipment_knowledge_lookup(query->available_equipment_id) == NULL)
+    return TRAINLOG_STATUS_NOT_FOUND;
+  for (i=0; i<trainlog_exercise_knowledge_count(); ++i) { const TrainlogExerciseKnowledge *e=&exercises[i];
+    const TrainlogKnowledgeInterpretation *v=e->interpretation; bool muscle_ok=true;
+    if (v == NULL) continue;
+    if (query->scientific_zone_id != NULL && !zone_matches(v,query->scientific_zone_id,query->include_zone_descendants)) continue;
+    if (query->movement_pattern_id != NULL && !list_has(v->pattern_ids,query->movement_pattern_id)) continue;
+    if (query->available_equipment_id != NULL && !list_has(e->equipment_ids,query->available_equipment_id)) continue;
+    if (query->muscle_id != NULL) {
+      switch (query->muscle_role) {
+        case TRAINLOG_KNOWLEDGE_ROLE_PRIMARY: muscle_ok=list_has(v->primary_muscle_ids,query->muscle_id); break;
+        case TRAINLOG_KNOWLEDGE_ROLE_SECONDARY: muscle_ok=list_has(v->secondary_muscle_ids,query->muscle_id); break;
+        case TRAINLOG_KNOWLEDGE_ROLE_STABILIZER: muscle_ok=list_has(v->stabilizer_muscle_ids,query->muscle_id); break;
+        case TRAINLOG_KNOWLEDGE_ROLE_ANY: muscle_ok=list_has(v->primary_muscle_ids,query->muscle_id) ||
+          list_has(v->secondary_muscle_ids,query->muscle_id) || list_has(v->stabilizer_muscle_ids,query->muscle_id); break;
+      }
+    }
+    if (!muscle_ok) continue;
+    if (count < capacity) output[count]=e;
+    ++count;
+  }
+  *output_count=count; return count > capacity ? TRAINLOG_STATUS_INVALID_ARGUMENT : TRAINLOG_STATUS_OK;
+}
+''')
+    output.write_text("\n".join(lines) + "\n", encoding="utf-8")
+
+
+if __name__ == "__main__":
+    main()

tools/generate_training_knowledge_audit.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""Regenerate the review audit solely from canonical TRAINING KNOWLEDGE V1."""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+
+def rows_from_catalog(catalog: Path) -> list[dict]:
+    exercises = json.loads((catalog / "exercise-knowledge-v1.json").read_text(encoding="utf-8"))["exercises"]
+    rows = []
+    for exercise in exercises:
+        authored = dict(exercise["body_zone_audit"])
+        interpretation = exercise["interpretation"] or exercise["conditional_interpretation"]
+        authored.update({
+            "exercise_id": exercise["exercise_id"],
+            "exercise_name": exercise["exercise_name"],
+            "resolution_status": exercise["resolution_status"],
+            "equipment_ids": exercise["equipment_ids"],
+            "knowledge_confidence": exercise["confidence"],
+            "knowledge_source_refs": exercise["source_refs"],
+            "family_description": None if interpretation is None else interpretation["family_description"],
+            "action_ids": [] if interpretation is None else interpretation["action_ids"],
+            "pattern_ids": [] if interpretation is None else interpretation["pattern_ids"],
+            "primary_muscle_ids": [] if interpretation is None else interpretation["primary_muscle_ids"],
+            "secondary_muscle_ids": [] if interpretation is None else interpretation["secondary_muscle_ids"],
+            "stabilizer_muscle_ids": [] if interpretation is None else interpretation["stabilizer_muscle_ids"],
+            "scientific_primary_zone_id": None if interpretation is None else interpretation["primary_zone_id"],
+            "scientific_secondary_zone_ids": [] if interpretation is None else interpretation["secondary_zone_ids"],
+        })
+        rows.append(authored)
+    return rows
+
+
+def generate(catalog: Path, output: Path) -> None:
+    rows = rows_from_catalog(catalog)
+    output.write_text(json.dumps({"format": "trainlog-training-knowledge-audit-v1",
+        "version": 1, "rows": rows}, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+
+
+def markdown_cell(value) -> str:
+    if value is None:
+        return "—"
+    if isinstance(value, list):
+        value = ", ".join(value) if value else "—"
+    return str(value).replace("|", "\\|").replace("\n", " ")
+
+
+def generate_markdown(catalog: Path, output: Path) -> None:
+    columns = (("exercise_id", "ID"), ("exercise_name", "Name"), ("action_ids", "Actions"),
+        ("pattern_ids", "Patterns"), ("primary_muscle_ids", "Primary"),
+        ("secondary_muscle_ids", "Secondary"), ("stabilizer_muscle_ids", "Stabilizers"),
+        ("scientific_primary_zone_id", "Science primary zone"),
+        ("scientific_secondary_zone_ids", "Science secondary zones"),
+        ("existing_primary_zone_id", "Existing primary zone"),
+        ("existing_secondary_zone_ids", "Existing secondary zones"), ("equipment_ids", "Equipment"),
+        ("knowledge_confidence", "Confidence"), ("knowledge_source_refs", "Source refs"),
+        ("status", "Audit status"))
+    lines = ["# Training knowledge science audit", "",
+        "Generated deterministically from the six canonical knowledge catalogs. Conditional rows are candidates that require explicit confirmation and do not participate in ordinary resolved queries.", "",
+        "| " + " | ".join(label for _, label in columns) + " |",
+        "| " + " | ".join("---" for _ in columns) + " |"]
+    for row in rows_from_catalog(catalog):
+        lines.append("| " + " | ".join(markdown_cell(row[key]) for key, _ in columns) + " |")
+    output.write_text("\n".join(lines) + "\n", encoding="utf-8")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("catalog", type=Path)
+    parser.add_argument("output", type=Path)
+    parser.add_argument("--markdown", action="store_true")
+    args = parser.parse_args()
+    if args.markdown:
+        generate_markdown(args.catalog, args.output)
+    else:
+        generate(args.catalog, args.output)
+
+
+if __name__ == "__main__":
+    main()

tools/validate_json.py

@@ -4,12 +4,12 @@
 from __future__ import annotations
 
 import argparse
+from dataclasses import dataclass
 import json
 import math
 import re
 import sys
 import unicodedata
-from datetime import datetime
 from pathlib import Path
 from typing import Any
 
@@ -120,25 +120,67 @@ def normalize_exercise_name(name: str) -> str:
     return collapsed.casefold()
 
 
-def parse_timestamp(value: str, field_name: str) -> datetime:
-    """Parse a Trainlog timestamp while requiring an explicit UTC offset."""
-    candidate = value
-    if candidate.endswith("Z"):
-        candidate = candidate[:-1] + "+00:00"
+@dataclass(frozen=True)
+class TrainlogTimestamp:
+    """Exact comparable instant key; fraction has insignificant zeros removed."""
 
-    try:
-        parsed = datetime.fromisoformat(candidate)
-    except ValueError as exc:
-        raise TrainlogSemanticError(
-            f"{field_name}: invalid date-time: {value!r}"
-        ) from exc
+    utc_second: int
+    fraction: str
 
-    if parsed.utcoffset() is None:
-        raise TrainlogSemanticError(
-            f"{field_name}: UTC offset is required: {value!r}"
-        )
+    def __lt__(self, other: "TrainlogTimestamp") -> bool:
+        return (self.utc_second, self.fraction) < (other.utc_second, other.fraction)
 
-    return parsed
+    def __le__(self, other: "TrainlogTimestamp") -> bool:
+        return self == other or self < other
+
+
+TIMESTAMP_PATTERN = re.compile(
+    r"(?P<year>[0-9]{4})-(?P<month>[0-9]{2})-(?P<day>[0-9]{2})"
+    r"[Tt](?P<hour>[0-9]{2}):(?P<minute>[0-9]{2})"
+    r"(?::(?P<second>[0-9]{2})(?:\.(?P<fraction>[0-9]+))?)?"
+    r"(?P<zone>[Zz]|[+-][0-9]{2}:[0-9]{2})",
+    re.ASCII,
+)
+
+
+def _leap_year(year: int) -> bool:
+    return year % 4 == 0 and (year % 100 != 0 or year % 400 == 0)
+
+
+def _days_in_month(year: int, month: int) -> int:
+    if month == 2:
+        return 29 if _leap_year(year) else 28
+    return 30 if month in {4, 6, 9, 11} else 31
+
+
+def _day_number(year: int, month: int, day: int) -> int:
+    prior = year - 1
+    result = prior * 365 + prior // 4 - prior // 100 + prior // 400
+    result += sum(_days_in_month(year, item) for item in range(1, month))
+    return result + day - 1
+
+
+def parse_timestamp(value: str, field_name: str) -> TrainlogTimestamp:
+    """Parse the frozen Trainlog grammar without platform ISO extensions."""
+    match = TIMESTAMP_PATTERN.fullmatch(value) if isinstance(value, str) else None
+    if match is None:
+        raise TrainlogSemanticError(f"{field_name}: invalid date-time: {value!r}")
+    year, month, day, hour, minute = (
+        int(match[name]) for name in ("year", "month", "day", "hour", "minute")
+    )
+    second = int(match["second"] or "0")
+    if year < 1 or month not in range(1, 13) or day not in range(
+            1, _days_in_month(year, month) + 1) or hour > 23 or minute > 59 or second > 59:
+        raise TrainlogSemanticError(f"{field_name}: invalid date-time: {value!r}")
+    zone = match["zone"]
+    offset = 0
+    if zone not in {"Z", "z"}:
+        offset_hour, offset_minute = int(zone[1:3]), int(zone[4:6])
+        if offset_hour > 23 or offset_minute > 59:
+            raise TrainlogSemanticError(f"{field_name}: invalid date-time: {value!r}")
+        offset = (offset_hour * 3600 + offset_minute * 60) * (1 if zone[0] == "+" else -1)
+    local = _day_number(year, month, day) * 86400 + hour * 3600 + minute * 60 + second
+    return TrainlogTimestamp(local - offset, (match["fraction"] or "").rstrip("0"))
 
 
 def require_non_blank(value: str, field_name: str) -> None:
@@ -358,8 +400,23 @@ def structural_errors(
     document: Any,
 ) -> list[str]:
     """Return deterministic human-readable JSON Schema errors."""
+    # WHY: jsonschema's optional RFC checker requires seconds, while Trainlog's
+    # existing Android writer may omit them. Admission must not depend on which
+    # optional dependencies are installed. Keep other supplied format checks,
+    # and override only date-time on a fresh instance; never mutate global state.
+    checker = jsonschema.FormatChecker()
+    if validator.format_checker is not None:
+        checker.checkers = validator.format_checker.checkers.copy()
+
+    @checker.checks("date-time", raises=TrainlogSemanticError)
+    def trainlog_date_time(value: Any) -> bool:
+        if isinstance(value, str):
+            parse_timestamp(value, "date-time")
+        return True  # JSON Schema's type keyword handles non-string values.
+
+    temporal_validator = validator.evolve(format_checker=checker)
     errors = sorted(
-        validator.iter_errors(document),
+        temporal_validator.iter_errors(document),
         key=lambda error: [str(part) for part in error.absolute_path],
     )
 
@@ -485,7 +542,15 @@ def main(argv: list[str]) -> int:
     validator.check_schema(schema)
 
     if not args.paths:
-        return run_suite(validator)
+        result = run_suite(validator)
+        try:
+            from validate_training_knowledge import ValidationError, validate as validate_knowledge
+            validate_knowledge(ROOT / "catalog")
+            print(f"PASS training-knowledge catalogs: {ROOT / 'catalog'}")
+        except ValidationError as error:
+            print(f"FAIL training-knowledge catalogs: {error}")
+            result = 1
+        return result
 
     failed = False
     for path in args.paths:

tools/validate_training_knowledge.py

@@ -0,0 +1,399 @@
+#!/usr/bin/env python3
+"""Strict cross-catalog validation for TRAINING KNOWLEDGE V1."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+from pathlib import Path
+
+
+class ValidationError(ValueError):
+    pass
+
+
+def unique_object(pairs):
+    result = {}
+    for key, value in pairs:
+        if key in result:
+            raise ValidationError(f"duplicate JSON key: {key}")
+        result[key] = value
+    return result
+
+
+def load(path: Path):
+    try:
+        with path.open(encoding="utf-8") as handle:
+            return json.load(handle, object_pairs_hook=unique_object)
+    except (OSError, json.JSONDecodeError) as error:
+        raise ValidationError(f"{path}: {error}") from error
+
+
+def require(condition: bool, message: str) -> None:
+    if not condition:
+        raise ValidationError(message)
+
+
+def strings(value, where: str, *, nonempty: bool = True, ordered: bool = True) -> None:
+    require(isinstance(value, list), f"{where}: expected array")
+    require(all(isinstance(item, str) and (item.strip() or not nonempty) for item in value),
+            f"{where}: expected strings")
+    require(len(value) == len(set(value)), f"{where}: duplicate values")
+    if ordered:
+        require(value == sorted(value), f"{where}: values must be sorted")
+
+
+def text(value, where: str) -> None:
+    require(isinstance(value, str) and bool(value.strip()), f"{where}: expected non-empty string")
+
+
+def exact_object(value, required: set[str], where: str, optional: set[str] | None = None) -> None:
+    optional = optional or set()
+    require(isinstance(value, dict), f"{where}: expected object")
+    require(required <= set(value) <= required | optional, f"{where}: invalid keys")
+
+
+def nullable_text(value, where: str) -> None:
+    require(value is None or isinstance(value, str), f"{where}: expected string or null")
+
+
+def confidence(value, where: str) -> None:
+    require(isinstance(value, str) and value in {"high", "moderate", "uncertain"},
+            f"{where}: invalid confidence")
+
+
+def refs_list(value, where: str, refs: set[str], *, nonempty: bool = True) -> None:
+    strings(value, where)
+    require(set(value) <= refs and (bool(value) or not nonempty), f"{where}: invalid references")
+
+
+def envelope(root, fmt: str, key: str):
+    require(set(root) == {"format", "version", key}, f"{key}: invalid envelope")
+    require(root["format"] == fmt and root["version"] == 1 and not isinstance(root["version"], bool),
+            f"{key}: unsupported version")
+    require(isinstance(root[key], list), f"{key}: expected array")
+    return root[key]
+
+
+def ids(rows, key: str, where: str) -> set[str]:
+    values = []
+    for row in rows:
+        require(isinstance(row, dict), f"{where}: record must be object")
+        text(row.get(key), f"{where}.{key}")
+        values.append(row[key])
+    require(values == sorted(values), f"{where}: records must be ordered by {key}")
+    require(len(values) == len(set(values)), f"{where}: duplicate {key}")
+    return set(values)
+
+
+def validate_interpretation(value, where, refs, muscles, actions, patterns, zones):
+    required = {"family_description", "action_ids", "pattern_ids", "primary_muscle_ids",
+                "secondary_muscle_ids", "stabilizer_muscle_ids", "primary_zone_id",
+                "secondary_zone_ids", "confidence", "evidence_type", "source_refs",
+                "variant_notes", "role_notes"}
+    exact_object(value, required, where, {"required_confirmation"})
+    for key in ("action_ids", "pattern_ids", "primary_muscle_ids", "secondary_muscle_ids",
+                "stabilizer_muscle_ids", "secondary_zone_ids", "source_refs"):
+        strings(value[key], f"{where}.{key}")
+    require(set(value["action_ids"]) <= actions, f"{where}: dangling action")
+    require(set(value["pattern_ids"]) <= patterns, f"{where}: dangling pattern")
+    role_lists = [set(value[key]) for key in ("primary_muscle_ids", "secondary_muscle_ids", "stabilizer_muscle_ids")]
+    require(set().union(*role_lists) <= muscles, f"{where}: dangling muscle")
+    require(sum(map(len, role_lists)) == len(set().union(*role_lists)), f"{where}: muscle appears in multiple roles")
+    require(value["primary_zone_id"] in zones and set(value["secondary_zone_ids"]) <= zones,
+            f"{where}: dangling zone")
+    require(value["primary_zone_id"] not in value["secondary_zone_ids"], f"{where}: duplicate zone role")
+    confidence(value["confidence"], f"{where}.confidence")
+    require(bool(value["source_refs"]) and set(value["source_refs"]) <= refs, f"{where}: dangling evidence")
+    for key in ("family_description", "evidence_type", "variant_notes", "role_notes"):
+        text(value[key], f"{where}.{key}")
+    if "required_confirmation" in value:
+        text(value["required_confirmation"], f"{where}.required_confirmation")
+
+
+def validate(root: Path) -> None:
+    body = load(root / "body-zones-v1.json")
+    require(body.get("format") == "trainlog-body-zone-catalog" and body.get("version") == 1,
+            "unsupported body-zone catalog")
+    require(all(row.get("kind") in {"group", "leaf", "standalone"} for row in body.get("zones", [])),
+            "invalid body-zone kind")
+    zones = {row["zone_id"] for row in body["zones"]}
+    references = envelope(load(root / "science-references-v1.json"), "trainlog-science-references-v1", "references")
+    muscles_rows = envelope(load(root / "muscles-v1.json"), "trainlog-muscles-v1", "muscles")
+    actions_rows = envelope(load(root / "joint-actions-v1.json"), "trainlog-joint-actions-v1", "joint_actions")
+    patterns_rows = envelope(load(root / "movement-patterns-v1.json"), "trainlog-movement-patterns-v1", "movement_patterns")
+    exercise_rows = envelope(load(root / "exercise-knowledge-v1.json"), "trainlog-exercise-knowledge-v1", "exercises")
+    equipment_rows = envelope(load(root / "equipment-knowledge-v1.json"), "trainlog-equipment-knowledge-v1", "equipment")
+    refs = ids(references, "ref_id", "references")
+    muscles = ids(muscles_rows, "muscle_id", "muscles")
+    actions = ids(actions_rows, "action_id", "joint_actions")
+    patterns = ids(patterns_rows, "pattern_id", "movement_patterns")
+    exercises = ids(exercise_rows, "exercise_id", "exercises")
+    equipment = ids(equipment_rows, "equipment_id", "equipment")
+    snake_id = re.compile(r"[a-z][a-z0-9_]*")
+    equipment_id = re.compile(r"(?:[a-z][a-z0-9_]*|eq_[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12})")
+    require(all(snake_id.fullmatch(value) for value in refs | muscles | actions | patterns),
+            "invalid stable knowledge ID syntax")
+    require(all(equipment_id.fullmatch(value) for value in equipment), "invalid stable equipment ID syntax")
+    equipment_manifest = load(root / "equipment-v1.json")
+    require(isinstance(equipment_manifest, dict) and
+            set(equipment_manifest) == {"format", "version", "equipment", "exercise_equipment"} and
+            equipment_manifest["format"] == "trainlog-equipment-catalog" and
+            equipment_manifest["version"] == 1, "invalid equipment manifest")
+    manifest_equipment = equipment_manifest["equipment"]
+    require(isinstance(manifest_equipment, list), "equipment manifest rows invalid")
+    required_equipment = set()
+    for manifest_row in manifest_equipment:
+        require(isinstance(manifest_row, dict), "equipment manifest row invalid")
+        text(manifest_row.get("id"), "equipment manifest id")
+        require(manifest_row["id"] not in required_equipment, "duplicate equipment manifest id")
+        required_equipment.add(manifest_row["id"])
+    mappings = body.get("exercise_mappings")
+    require(isinstance(mappings, list), "body-zone exercise mappings invalid")
+    required_exercises = set()
+    for mapping in mappings:
+        require(isinstance(mapping, dict), "body-zone exercise mapping invalid")
+        text(mapping.get("exercise_id"), "body-zone exercise mapping id")
+        require(mapping["exercise_id"] not in required_exercises, "duplicate body-zone exercise mapping")
+        required_exercises.add(mapping["exercise_id"])
+    require(required_equipment <= equipment, "knowledge catalog misses supplied equipment")
+    require(required_exercises <= exercises, "knowledge catalog misses observed exercises")
+
+    ref_required = {"ref_id", "title", "authors_or_organization", "year", "type", "url", "topics",
+                    "notes", "limitations", "doi", "pmid", "accessed_on"}
+    for row in references:
+        exact_object(row, ref_required, "reference", {"publication_note"})
+        for key in ("title", "authors_or_organization", "type", "url", "notes", "limitations", "accessed_on"):
+            text(row[key], f"reference.{key}")
+        require(row["year"] is None or isinstance(row["year"], int) and not isinstance(row["year"], bool),
+                "reference.year invalid")
+        strings(row["topics"], "reference.topics", ordered=False)
+        require(row["doi"] is None or isinstance(row["doi"], str), "reference.doi invalid")
+        require(row["pmid"] is None or isinstance(row["pmid"], str), "reference.pmid invalid")
+        if "publication_note" in row:
+            nullable_text(row["publication_note"], "reference.publication_note")
+    ref_types = {row["ref_id"]: row["type"] for row in references}
+    scientific_types = {"established_anatomy", "emg_evidence", "intervention_evidence"}
+
+    for row in muscles_rows:
+        required = {"muscle_id", "display_name", "display_name_fr", "entity_type", "anatomical_group",
+                    "aggregate_group_id", "body_zone_id", "body_zone_ids", "joint_action_ids",
+                    "primary_actions", "primary_actions_semantics", "overlap_warning", "functional_notes",
+                    "confidence", "evidence_type", "source_refs"}
+        exact_object(row, required, f"muscle {row['muscle_id']}", {"member_muscle_ids"})
+        for key in ("display_name", "display_name_fr", "anatomical_group", "primary_actions_semantics",
+                    "overlap_warning", "evidence_type"):
+            text(row[key], f"muscle.{key}")
+        require(isinstance(row["functional_notes"], str), "muscle.functional_notes invalid")
+        require(row["entity_type"] in {"muscle", "muscle_region", "muscle_group"}, "invalid muscle entity_type")
+        require(row["aggregate_group_id"] is None or row["aggregate_group_id"] in muscles, "dangling aggregate group")
+        if "member_muscle_ids" in row:
+            strings(row["member_muscle_ids"], "muscle.member_muscle_ids")
+            require(set(row["member_muscle_ids"]) <= muscles, "dangling group member")
+        require(row["body_zone_id"] in zones, "dangling muscle zone")
+        for key in ("body_zone_ids", "joint_action_ids", "primary_actions", "source_refs"):
+            strings(row[key], f"muscle.{key}")
+        require(set(row["body_zone_ids"]) <= zones and set(row["joint_action_ids"]) <= actions and
+                set(row["primary_actions"]) <= actions and set(row["source_refs"]) <= refs and row["source_refs"],
+                "muscle cross-reference failure")
+        confidence(row["confidence"], "muscle.confidence")
+
+    for row in actions_rows:
+        required = {"action_id", "display_name_fr", "definition", "anatomical_region", "joint_complex",
+                    "joint_or_complex", "principal_plane", "plane_notes", "contributing_muscle_ids",
+                    "contributor_semantics", "evidence_type", "confidence", "source_refs", "notes"}
+        require(set(row) == required, f"action {row['action_id']}: invalid keys")
+        for key in ("display_name_fr", "definition", "anatomical_region", "joint_complex", "joint_or_complex",
+                    "principal_plane", "plane_notes", "contributor_semantics", "evidence_type", "notes"):
+            text(row[key], f"action.{key}")
+        strings(row["contributing_muscle_ids"], "action.contributing_muscle_ids")
+        strings(row["source_refs"], "action.source_refs")
+        require(set(row["contributing_muscle_ids"]) <= muscles and set(row["source_refs"]) <= refs and row["source_refs"],
+                "action cross-reference failure")
+        confidence(row["confidence"], "action.confidence")
+
+    for row in patterns_rows:
+        required = {"pattern_id", "display_name_fr", "definition", "typical_action_ids", "typical_body_zone_ids",
+                    "evidence_type", "confidence", "source_refs", "notes"}
+        require(required <= set(row) <= required | {"body_zone_semantics", "parent_pattern_id"},
+                f"pattern {row['pattern_id']}: invalid keys")
+        for key in ("display_name_fr", "definition", "evidence_type", "notes"):
+            text(row[key], f"pattern.{key}")
+        for key in ("body_zone_semantics", "parent_pattern_id"):
+            if key in row:
+                nullable_text(row[key], f"pattern.{key}")
+        for key in ("typical_action_ids", "typical_body_zone_ids", "source_refs"):
+            strings(row[key], f"pattern.{key}")
+        require(set(row["typical_action_ids"]) <= actions and set(row["typical_body_zone_ids"]) <= zones and
+                set(row["source_refs"]) <= refs and row["source_refs"], "pattern cross-reference failure")
+        if row.get("parent_pattern_id") is not None:
+            require(row["parent_pattern_id"] in patterns, "dangling parent pattern")
+        confidence(row["confidence"], "pattern.confidence")
+
+    linked = set()
+    exercise_equipment = {}
+    for row in exercise_rows:
+        required = {"exercise_id", "exercise_name", "equipment_ids", "identity_evidence",
+                    "identity_evidence_type", "resolution_status", "confidence", "interpretation",
+                    "conditional_interpretation", "existing_body_zones", "source_refs", "limitations",
+                    "body_zone_audit"}
+        require(required <= set(row) <= required | {"equipment_link_status"},
+                f"exercise {row['exercise_id']}: invalid keys")
+        require(re.fullmatch(r"ex_[0-9a-f]{8}-[0-9a-f]{4}-4[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}",
+                             row["exercise_id"]) is not None, "invalid exercise UUID")
+        for key in ("exercise_name", "identity_evidence", "identity_evidence_type"):
+            text(row[key], f"exercise.{key}")
+        strings(row["limitations"], "exercise.limitations", ordered=False)
+        if "equipment_link_status" in row:
+            text(row["equipment_link_status"], "exercise.equipment_link_status")
+        exact_object(row["existing_body_zones"], {"primary_zone_id", "secondary_zone_ids"},
+                     "exercise.existing_body_zones")
+        require(row["existing_body_zones"]["primary_zone_id"] is None or
+                row["existing_body_zones"]["primary_zone_id"] in zones,
+                "exercise existing primary zone invalid")
+        strings(row["existing_body_zones"]["secondary_zone_ids"], "exercise.existing_body_zones.secondary_zone_ids")
+        require(set(row["existing_body_zones"]["secondary_zone_ids"]) <= zones,
+                "exercise existing secondary zones invalid")
+        body_audit = row["body_zone_audit"]
+        exact_object(body_audit, {"status", "severity", "confidence", "rationale", "existing_primary_zone_id",
+                    "existing_secondary_zone_ids", "proposed_mutation", "source_refs"}, "exercise.body_zone_audit")
+        require(body_audit["status"] in {"confirmed", "questionable", "unresolved"}, "invalid embedded body-zone audit")
+        for key in ("severity", "rationale"):
+            text(body_audit[key], f"exercise.body_zone_audit.{key}")
+        confidence(body_audit["confidence"], "exercise.body_zone_audit.confidence")
+        require(body_audit["existing_primary_zone_id"] is None or body_audit["existing_primary_zone_id"] in zones,
+                "audit primary zone invalid")
+        strings(body_audit["existing_secondary_zone_ids"], "audit existing secondary zones")
+        require(set(body_audit["existing_secondary_zone_ids"]) <= zones, "audit secondary zone invalid")
+        nullable_text(body_audit["proposed_mutation"], "audit.proposed_mutation")
+        refs_list(body_audit["source_refs"], "exercise.body_zone_audit.source_refs", refs,
+                  nonempty=body_audit["status"] != "unresolved")
+        status = row["resolution_status"]
+        require(status in {"resolved_family_variant_limited", "conditional", "unresolved"}, "invalid resolution status")
+        confidence(row["confidence"], "exercise.confidence")
+        strings(row["equipment_ids"], "exercise.equipment_ids")
+        exercise_equipment[row["exercise_id"]] = set(row["equipment_ids"])
+        strings(row["source_refs"], "exercise.source_refs")
+        require(set(row["equipment_ids"]) <= equipment and set(row["source_refs"]) <= refs,
+                "exercise cross-reference failure")
+        if status == "resolved_family_variant_limited":
+            require(bool(row["source_refs"]), "resolved exercise lacks evidence")
+            require(row["conditional_interpretation"] is None, "resolved record has conditional interpretation")
+            validate_interpretation(row["interpretation"], f"exercise {row['exercise_id']}", refs, muscles, actions, patterns, zones)
+            if row["confidence"] == "high":
+                require(any(ref_types[ref] in scientific_types for ref in row["interpretation"]["source_refs"]),
+                        "high exercise mapping lacks scientific source")
+        elif status == "conditional":
+            require(bool(row["source_refs"]), "conditional exercise lacks evidence")
+            require(row["interpretation"] is None, "conditional record leaks into resolved interpretation")
+            validate_interpretation(row["conditional_interpretation"], f"conditional {row['exercise_id']}", refs, muscles, actions, patterns, zones)
+        else:
+            require(row["interpretation"] is None and row["conditional_interpretation"] is None,
+                    "unresolved record has interpretation")
+
+    manufacturer_only = {"manufacturer_statement"}
+    equipment_required = {"equipment_id", "manufacturer", "model", "identification_status", "scientific_status",
+        "scientific_status_scope", "catalog_type", "catalog_load_semantics", "mechanics", "evidence_type",
+        "confidence", "source_refs", "capabilities", "requires_actual_exercise", "limitations"}
+    capability_required = {"display_name", "exercise_ids", "link_status", "interpretation", "requirements"}
+    for row in equipment_rows:
+        exact_object(row, equipment_required, f"equipment {row['equipment_id']}", {"audit_note"})
+        for key in ("identification_status", "scientific_status_scope", "catalog_type",
+                    "mechanics", "evidence_type"):
+            text(row[key], f"equipment.{key}")
+        for key in ("manufacturer", "model", "catalog_load_semantics"):
+            nullable_text(row[key], f"equipment.{key}")
+        if "audit_note" in row:
+            nullable_text(row["audit_note"], "equipment.audit_note")
+        require(isinstance(row["requires_actual_exercise"], bool), "equipment requires_actual_exercise invalid")
+        strings(row["limitations"], "equipment.limitations", ordered=False)
+        require(row["scientific_status"] in {"scientifically_documented", "mechanically_identified_anatomy_incomplete",
+                                         "equipment_identity_uncertain"}, "invalid equipment science_status")
+        require(row["confidence"] in {"high", "moderate", "uncertain"}, "invalid equipment confidence")
+        strings(row["source_refs"], "equipment.source_refs")
+        require(set(row["source_refs"]) <= refs and row["source_refs"], "equipment evidence failure")
+        if row["confidence"] == "high":
+            require(row["evidence_type"] not in manufacturer_only, "manufacturer-only high anatomical mapping")
+            require(any(ref_types[ref] in scientific_types for ref in row["source_refs"]),
+                    "high equipment mapping lacks scientific source")
+        require(isinstance(row["capabilities"], list), "equipment capabilities invalid")
+        for capability in row["capabilities"]:
+            exact_object(capability, capability_required, "equipment capability")
+            for key in ("display_name", "link_status", "requirements"):
+                text(capability[key], f"capability.{key}")
+            strings(capability["exercise_ids"], "capability.exercise_ids")
+            require(set(capability["exercise_ids"]) <= exercises, "capability phantom exercise")
+            linked.update((exercise_id, row["equipment_id"]) for exercise_id in capability["exercise_ids"])
+            for exercise_id in capability["exercise_ids"]:
+                require(row["equipment_id"] in exercise_equipment[exercise_id],
+                        "equipment/exercise compatibility is not symmetric")
+            if capability["interpretation"] is not None:
+                validate_interpretation(capability["interpretation"], "equipment capability", refs, muscles, actions, patterns, zones)
+    for row in exercise_rows:
+        for equipment_id in row["equipment_ids"]:
+            require((row["exercise_id"], equipment_id) in linked, "exercise/equipment compatibility is not symmetric")
+
+    audit = envelope(load(root / "training-knowledge-audit-v1.json"),
+                     "trainlog-training-knowledge-audit-v1", "rows")
+    require(ids(audit, "exercise_id", "audit") == exercises, "audit does not cover exercise inventory")
+    for row in audit:
+        audit_required = {"status", "severity", "confidence", "rationale", "existing_primary_zone_id",
+            "existing_secondary_zone_ids", "proposed_mutation", "source_refs", "exercise_id", "exercise_name",
+            "resolution_status", "equipment_ids", "knowledge_confidence", "knowledge_source_refs",
+            "family_description", "action_ids", "pattern_ids", "primary_muscle_ids", "secondary_muscle_ids",
+            "stabilizer_muscle_ids", "scientific_primary_zone_id", "scientific_secondary_zone_ids"}
+        exact_object(row, audit_required, "audit row")
+        require(row["status"] in {"confirmed", "questionable", "unresolved"}, "invalid audit status")
+        confidence(row["confidence"], "audit.confidence")
+        confidence(row["knowledge_confidence"], "audit.knowledge_confidence")
+        for key in ("exercise_name", "resolution_status", "severity", "rationale"):
+            text(row[key], f"audit.{key}")
+        for key in ("family_description", "scientific_primary_zone_id", "proposed_mutation"):
+            nullable_text(row[key], f"audit.{key}")
+        for key in ("source_refs", "knowledge_source_refs", "equipment_ids", "action_ids", "pattern_ids",
+                    "primary_muscle_ids", "secondary_muscle_ids", "stabilizer_muscle_ids",
+                    "scientific_secondary_zone_ids"):
+            strings(row[key], f"audit.{key}")
+        require(set(row["source_refs"]) <= refs and set(row["knowledge_source_refs"]) <= refs,
+                "audit evidence failure")
+        require(set(row["equipment_ids"]) <= equipment and set(row["action_ids"]) <= actions and
+                set(row["pattern_ids"]) <= patterns and
+                set(row["primary_muscle_ids"] + row["secondary_muscle_ids"] +
+                    row["stabilizer_muscle_ids"]) <= muscles and
+                set(row["scientific_secondary_zone_ids"]) <= zones,
+                "audit cross-reference failure")
+        require(row["status"] == "unresolved" or bool(row["source_refs"]), "resolved audit lacks evidence")
+    expected_audit = []
+    for exercise in exercise_rows:
+        expected = dict(exercise["body_zone_audit"])
+        interpretation = exercise["interpretation"] or exercise["conditional_interpretation"]
+        expected.update({
+            "exercise_id": exercise["exercise_id"], "exercise_name": exercise["exercise_name"],
+            "resolution_status": exercise["resolution_status"], "equipment_ids": exercise["equipment_ids"],
+            "knowledge_confidence": exercise["confidence"], "knowledge_source_refs": exercise["source_refs"],
+            "family_description": None if interpretation is None else interpretation["family_description"],
+            "action_ids": [] if interpretation is None else interpretation["action_ids"],
+            "pattern_ids": [] if interpretation is None else interpretation["pattern_ids"],
+            "primary_muscle_ids": [] if interpretation is None else interpretation["primary_muscle_ids"],
+            "secondary_muscle_ids": [] if interpretation is None else interpretation["secondary_muscle_ids"],
+            "stabilizer_muscle_ids": [] if interpretation is None else interpretation["stabilizer_muscle_ids"],
+            "scientific_primary_zone_id": None if interpretation is None else interpretation["primary_zone_id"],
+            "scientific_secondary_zone_ids": [] if interpretation is None else interpretation["secondary_zone_ids"],
+        })
+        expected_audit.append(expected)
+    require(audit == expected_audit, "generated audit is stale or inconsistent with canonical exercises")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("catalog", nargs="?", type=Path, default=Path("catalog"))
+    args = parser.parse_args()
+    try:
+        validate(args.catalog)
+    except ValidationError as error:
+        raise SystemExit(str(error)) from error
+
+
+if __name__ == "__main__":
+    main()

tui/include/trainlog/database.h

@@ -15,6 +15,11 @@
 
 typedef struct TrainlogDatabase TrainlogDatabase;
 
+/* Nestable read savepoints let composition services observe one database state
+ * without exposing SQLite or issuing writes to application data. */
+TrainlogStatus trainlog_database_read_snapshot_begin(TrainlogDatabase *database);
+TrainlogStatus trainlog_database_read_snapshot_end(TrainlogDatabase *database, bool commit_snapshot);
+
 /* WHY: session occurrences retain an equipment ID, while custom definitions
  * need durable presentation metadata. A reference alone is never a definition. */
 typedef struct TrainlogCustomEquipment {
@@ -386,6 +391,115 @@ TrainlogStatus trainlog_database_list_exercise_performance(
     size_t *output_count
 );
 
+/* TRAINING_KNOWLEDGE_RUNTIME_READ_V1 */
+#define TRAINLOG_OCCURRENCE_PAGE_MAX 32U
+#define TRAINLOG_OCCURRENCE_SET_PAGE_MAX 64U
+
+typedef struct TrainlogExerciseOccurrenceCursor {
+    char started_at[TRAINLOG_TIMESTAMP_MAX + 1U];
+    char session_id[TRAINLOG_ID_MAX + 1U];
+    char entry_id[TRAINLOG_ID_MAX + 1U];
+} TrainlogExerciseOccurrenceCursor;
+
+typedef struct TrainlogExerciseOccurrence {
+    char session_id[TRAINLOG_ID_MAX + 1U];
+    char entry_id[TRAINLOG_ID_MAX + 1U];
+    char exercise_id[TRAINLOG_ID_MAX + 1U];
+    char started_at[TRAINLOG_TIMESTAMP_MAX + 1U];
+    char equipment_id[TRAINLOG_ID_MAX + 1U];
+    TrainlogSessionType session_type;
+    TrainlogTrackingMode tracking_mode;
+    TrainlogRecordingMode recording_mode;
+    TrainlogExerciseDataFields data_fields;
+    TrainlogLoadMode load_mode;
+    size_t set_count;
+    int continuous_duration_seconds;
+    bool continuous_has_speed;
+    double continuous_speed_kmh;
+    bool continuous_has_distance;
+    double continuous_distance_km;
+} TrainlogExerciseOccurrence;
+
+typedef struct TrainlogOccurrenceSet {
+    size_t position;
+    bool has_reps;
+    int reps;
+    bool has_duration;
+    int duration_seconds;
+    bool has_weight;
+    double weight_kg;
+} TrainlogOccurrenceSet;
+
+typedef struct TrainlogLatestExplicitMax {
+    bool found;
+    char session_id[TRAINLOG_ID_MAX + 1U];
+    char entry_id[TRAINLOG_ID_MAX + 1U];
+    char started_at[TRAINLOG_TIMESTAMP_MAX + 1U];
+    char equipment_id[TRAINLOG_ID_MAX + 1U];
+    TrainlogLoadMode load_mode;
+    double max_weight_kg;
+} TrainlogLatestExplicitMax;
+
+/* Exact-ID profile read; unknown IDs return NOT_FOUND. */
+TrainlogStatus trainlog_database_get_exercise_profile(
+    TrainlogDatabase *database,
+    const char *exercise_id,
+    TrainlogExercise *output
+);
+
+/**
+ * Current-data keyset page ordered by the exact started_at instant, then
+ * session_id and entry_id bytewise DESC. Accepted timestamps use the frozen
+ * extended RFC3339 grammar, including T/t, Z/z, numeric offsets through
+ * 23:59, arbitrary fractional precision, and the Android writer's omitted
+ * seconds form. Equivalent trailing-zero fractions represent one instant.
+ * A non-NULL cursor is exclusive. Pages are not a cross-call snapshot under
+ * concurrent edits. Cursor arrays are borrowed for this call, must be NUL
+ * terminated within their declared capacities, and started_at must be a real
+ * accepted instant with an explicit numeric offset or Z/z. Malformed cursors
+ * are INVALID_ARGUMENT before any history query. Malformed matching persisted
+ * timestamps, or a selected timestamp too long for the fixed public output,
+ * are DATABASE_ERROR rather than omitted or truncated. limit must be
+ * 1..TRAINLOG_OCCURRENCE_PAGE_MAX.
+ */
+TrainlogStatus trainlog_database_list_exercise_occurrences_page(
+    TrainlogDatabase *database,
+    const char *exercise_id,
+    const TrainlogExerciseOccurrenceCursor *after,
+    size_t limit,
+    TrainlogExerciseOccurrence *output,
+    size_t *output_count,
+    bool *output_has_more,
+    TrainlogExerciseOccurrenceCursor *output_next
+);
+
+/* Original set positions are preserved. after_position is exclusive; use -1
+ * for the first page. Reps are non-negative, so zero preserves a failed
+ * attempt. Nullable values remain distinct from zero. Corrupt storage types or
+ * values outside the public int/size_t ranges return DATABASE_ERROR. No rows
+ * exist for continuous occurrences. Output remains caller-owned. */
+TrainlogStatus trainlog_database_list_occurrence_sets_page(
+    TrainlogDatabase *database,
+    const char *entry_id,
+    int after_position,
+    size_t limit,
+    TrainlogOccurrenceSet *output,
+    size_t *output_count,
+    bool *output_has_more,
+    int *output_next_position
+);
+
+/* Reads max_results joined only through max_test sessions; ordinary large sets
+ * never qualify. It uses the same exact instant and bytewise-ID ordering as
+ * occurrence pagination. Output keeps the exact source timestamp and
+ * occurrence load/equipment context; corrupt or over-capacity selected storage
+ * returns DATABASE_ERROR. */
+TrainlogStatus trainlog_database_latest_explicit_max_context(
+    TrainlogDatabase *database,
+    const char *exercise_id,
+    TrainlogLatestExplicitMax *output
+);
+
 
 /* TRAINLOG_SESSION_EDIT_API */
 

tui/include/trainlog/training_context.h

@@ -0,0 +1,50 @@
+#ifndef TRAINLOG_TRAINING_CONTEXT_H
+#define TRAINLOG_TRAINING_CONTEXT_H
+
+/** @file training_context.h Read-only runtime/scientific exercise composition. */
+
+#include "trainlog/database.h"
+#include "trainlog/training_knowledge.h"
+
+typedef struct TrainlogTrainingOccurrenceView {
+    TrainlogExerciseOccurrence occurrence;
+    size_t set_offset;
+    size_t set_count;
+    bool sets_have_more;
+    int next_set_position;
+} TrainlogTrainingOccurrenceView;
+
+typedef struct TrainlogTrainingExerciseContext {
+    TrainlogExercise exercise;
+    TrainlogExerciseBodyZone *persisted_zones;
+    size_t persisted_zone_count;
+    const TrainlogExerciseKnowledge *knowledge; /* borrowed process-lifetime data */
+    const TrainlogEquipmentKnowledge **compatible_equipment; /* owned pointer array */
+    size_t compatible_equipment_count;
+    TrainlogLatestExplicitMax latest_max;
+    TrainlogTrainingOccurrenceView *occurrences;
+    size_t occurrence_count;
+    bool occurrences_have_more;
+    TrainlogExerciseOccurrenceCursor next_occurrence;
+    TrainlogOccurrenceSet *sets;
+    size_t set_count;
+} TrainlogTrainingExerciseContext;
+
+/**
+ * Compose one exact runtime ID. occurrence_limit is 1..32 and set_preview_limit
+ * is 1..64 per occurrence. The caller owns output allocations and must call
+ * release after success. Unknown runtime IDs return NOT_FOUND; missing knowledge
+ * is valid and represented by NULL. One nested-safe database read snapshot
+ * covers the call. No catalog relationship fabricates occurrence history.
+ */
+TrainlogStatus trainlog_training_exercise_context_load(
+    TrainlogDatabase *database,
+    const char *exercise_id,
+    size_t occurrence_limit,
+    size_t set_preview_limit,
+    TrainlogTrainingExerciseContext *output
+);
+
+void trainlog_training_exercise_context_release(TrainlogTrainingExerciseContext *context);
+
+#endif

tui/include/trainlog/training_knowledge.h

@@ -0,0 +1,219 @@
+#ifndef TRAINLOG_TRAINING_KNOWLEDGE_H
+#define TRAINLOG_TRAINING_KNOWLEDGE_H
+
+/**
+ * @file training_knowledge.h
+ * @brief Immutable, evidence-linked TRAINING KNOWLEDGE V1 catalog API.
+ *
+ * All returned records and strings are borrowed from generated const storage
+ * and remain valid for process lifetime. Newline-separated ID fields contain
+ * exact stable IDs; labels are never used as identities.
+ */
+
+#include <stdbool.h>
+#include <stddef.h>
+
+#include "trainlog/status.h"
+
+typedef enum TrainlogKnowledgeMuscleRole {
+    TRAINLOG_KNOWLEDGE_ROLE_ANY = 0,
+    TRAINLOG_KNOWLEDGE_ROLE_PRIMARY,
+    TRAINLOG_KNOWLEDGE_ROLE_SECONDARY,
+    TRAINLOG_KNOWLEDGE_ROLE_STABILIZER
+} TrainlogKnowledgeMuscleRole;
+
+typedef struct TrainlogKnowledgeReference {
+    const char *ref_id;
+    const char *title;
+    const char *authors_or_organization;
+    int year; /* zero means the catalog explicitly records an unknown year */
+    const char *type;
+    const char *url;
+    const char *doi;
+    const char *pmid;
+    const char *topics;
+    const char *notes;
+    const char *limitations;
+    const char *accessed_on;
+    const char *publication_note;
+} TrainlogKnowledgeReference;
+
+typedef struct TrainlogKnowledgeMuscle {
+    const char *muscle_id;
+    const char *display_name;
+    const char *display_name_fr;
+    const char *entity_type;
+    const char *anatomical_group;
+    const char *aggregate_group_id;
+    const char *member_muscle_ids;
+    const char *body_zone_ids;
+    const char *joint_action_ids;
+    const char *primary_actions;
+    const char *primary_actions_semantics;
+    const char *functional_notes;
+    const char *confidence;
+    const char *evidence_type;
+    const char *source_refs;
+    const char *overlap_warning;
+} TrainlogKnowledgeMuscle;
+
+typedef struct TrainlogKnowledgeJointAction {
+    const char *action_id;
+    const char *display_name_fr;
+    const char *definition;
+    const char *anatomical_region;
+    const char *joint_complex;
+    const char *principal_plane;
+    const char *plane_notes;
+    const char *contributing_muscle_ids;
+    const char *contributor_semantics;
+    const char *confidence;
+    const char *evidence_type;
+    const char *source_refs;
+    const char *notes;
+} TrainlogKnowledgeJointAction;
+
+typedef struct TrainlogKnowledgeMovementPattern {
+    const char *pattern_id;
+    const char *display_name_fr;
+    const char *definition;
+    const char *parent_pattern_id;
+    const char *typical_action_ids;
+    const char *typical_body_zone_ids;
+    const char *body_zone_semantics;
+    const char *confidence;
+    const char *evidence_type;
+    const char *source_refs;
+    const char *notes;
+} TrainlogKnowledgeMovementPattern;
+
+typedef struct TrainlogKnowledgeInterpretation {
+    const char *family_description;
+    const char *action_ids;
+    const char *pattern_ids;
+    const char *primary_muscle_ids;
+    const char *secondary_muscle_ids;
+    const char *stabilizer_muscle_ids;
+    const char *primary_zone_id;
+    const char *secondary_zone_ids;
+    const char *confidence;
+    const char *evidence_type;
+    const char *source_refs;
+    const char *variant_notes;
+    const char *role_notes;
+    const char *required_confirmation;
+} TrainlogKnowledgeInterpretation;
+
+typedef struct TrainlogExerciseKnowledge {
+    const char *exercise_id;
+    const char *exercise_name;
+    const char *resolution_status;
+    const char *confidence;
+    const char *equipment_ids;
+    const char *identity_evidence;
+    const char *identity_evidence_type;
+    const char *source_refs;
+    const char *limitations;
+    const char *equipment_link_status;
+    const TrainlogKnowledgeInterpretation *interpretation;
+    const TrainlogKnowledgeInterpretation *conditional_interpretation;
+} TrainlogExerciseKnowledge;
+
+/* WHY: the authored BODY ZONES review is part of the immutable scientific
+ * catalog contract. Exposing every field prevents clients from substituting
+ * the resolved interpretation for the review decision. All pointers are
+ * borrowed from generated storage and remain valid for process lifetime. */
+typedef struct TrainlogKnowledgeBodyZoneAudit {
+    const char *exercise_id;
+    const char *status;
+    const char *severity;
+    const char *rationale;
+    const char *confidence;
+    const char *source_refs;
+    const char *existing_primary_zone_id;
+    const char *existing_secondary_zone_ids;
+    const char *scientific_primary_zone_id;
+    const char *scientific_secondary_zone_ids;
+    const char *proposed_mutation;
+} TrainlogKnowledgeBodyZoneAudit;
+
+typedef struct TrainlogEquipmentKnowledge {
+    const char *equipment_id;
+    const char *manufacturer;
+    const char *model;
+    const char *identification_status;
+    const char *scientific_status;
+    const char *scientific_status_scope;
+    const char *catalog_type;
+    const char *catalog_load_semantics;
+    const char *mechanics;
+    const char *confidence;
+    const char *evidence_type;
+    const char *source_refs;
+    const char *limitations;
+    const char *audit_note;
+    bool requires_actual_exercise;
+} TrainlogEquipmentKnowledge;
+
+typedef struct TrainlogEquipmentCapability {
+    const char *equipment_id;
+    const char *display_name;
+    const char *exercise_ids;
+    const char *link_status;
+    const char *requirements;
+    const TrainlogKnowledgeInterpretation *interpretation;
+} TrainlogEquipmentCapability;
+
+typedef struct TrainlogKnowledgeQuery {
+    const char *scientific_zone_id;
+    bool include_zone_descendants;
+    const char *movement_pattern_id;
+    const char *muscle_id;
+    TrainlogKnowledgeMuscleRole muscle_role;
+    const char *available_equipment_id;
+} TrainlogKnowledgeQuery;
+
+size_t trainlog_knowledge_reference_count(void);
+const TrainlogKnowledgeReference *trainlog_knowledge_reference_at(size_t index);
+const TrainlogKnowledgeReference *trainlog_knowledge_reference_lookup(const char *ref_id);
+size_t trainlog_knowledge_muscle_count(void);
+const TrainlogKnowledgeMuscle *trainlog_knowledge_muscle_at(size_t index);
+const TrainlogKnowledgeMuscle *trainlog_knowledge_muscle_lookup(const char *muscle_id);
+size_t trainlog_knowledge_joint_action_count(void);
+const TrainlogKnowledgeJointAction *trainlog_knowledge_joint_action_at(size_t index);
+const TrainlogKnowledgeJointAction *trainlog_knowledge_joint_action_lookup(const char *action_id);
+size_t trainlog_knowledge_movement_pattern_count(void);
+const TrainlogKnowledgeMovementPattern *trainlog_knowledge_movement_pattern_at(size_t index);
+const TrainlogKnowledgeMovementPattern *trainlog_knowledge_movement_pattern_lookup(const char *pattern_id);
+size_t trainlog_exercise_knowledge_count(void);
+const TrainlogExerciseKnowledge *trainlog_exercise_knowledge_at(size_t index);
+const TrainlogExerciseKnowledge *trainlog_exercise_knowledge_lookup(const char *exercise_id);
+/* at() returns NULL outside the immutable snapshot; lookup() returns NULL for
+ * NULL, empty, or unknown IDs. Neither function transfers ownership. */
+size_t trainlog_knowledge_body_zone_audit_count(void);
+const TrainlogKnowledgeBodyZoneAudit *trainlog_knowledge_body_zone_audit_at(size_t index);
+const TrainlogKnowledgeBodyZoneAudit *trainlog_knowledge_body_zone_audit_lookup(const char *exercise_id);
+/* Only this explicitly named accessor exposes conditional candidate data. */
+const TrainlogKnowledgeInterpretation *trainlog_exercise_knowledge_conditional(const char *exercise_id);
+size_t trainlog_equipment_knowledge_count(void);
+const TrainlogEquipmentKnowledge *trainlog_equipment_knowledge_at(size_t index);
+const TrainlogEquipmentKnowledge *trainlog_equipment_knowledge_lookup(const char *equipment_id);
+size_t trainlog_equipment_capability_count(void);
+const TrainlogEquipmentCapability *trainlog_equipment_capability_at(size_t index);
+
+/**
+ * AND-combine optional resolved-knowledge filters in stable exercise-ID order.
+ * muscle_id may be NULL only when muscle_role is ROLE_ANY; a specific role
+ * and muscle_id must be supplied together. Unknown filter IDs return NOT_FOUND. Malformed arguments or insufficient
+ * capacity return INVALID_ARGUMENT. output_count always receives the required
+ * count after valid filters are resolved, so truncation is never reported as
+ * success. Conditional and unresolved records can never match this function.
+ */
+TrainlogStatus trainlog_exercise_knowledge_query(
+    const TrainlogKnowledgeQuery *query,
+    const TrainlogExerciseKnowledge **output,
+    size_t capacity,
+    size_t *output_count
+);
+
+#endif

tui/meson.build

@@ -30,6 +30,28 @@ body_zone_catalog_generated = custom_target(
   command: [find_program('python3'), meson.project_source_root() / 'tools/generate_body_zone_catalog.py', '@INPUT@', '@OUTPUT@'],
 )
 
+training_knowledge_generated = custom_target(
+  'training_knowledge_generated',
+  input: [
+    meson.project_source_root() / 'catalog/body-zones-v1.json',
+    meson.project_source_root() / 'catalog/equipment-v1.json',
+    meson.project_source_root() / 'catalog/science-references-v1.json',
+    meson.project_source_root() / 'catalog/muscles-v1.json',
+    meson.project_source_root() / 'catalog/joint-actions-v1.json',
+    meson.project_source_root() / 'catalog/movement-patterns-v1.json',
+    meson.project_source_root() / 'catalog/exercise-knowledge-v1.json',
+    meson.project_source_root() / 'catalog/equipment-knowledge-v1.json',
+    meson.project_source_root() / 'catalog/training-knowledge-audit-v1.json',
+  ],
+  output: 'training_knowledge_generated.c',
+  depend_files: [
+    meson.project_source_root() / 'tools/generate_training_knowledge.py',
+    meson.project_source_root() / 'tools/validate_training_knowledge.py',
+  ],
+  command: [find_program('python3'), meson.project_source_root() / 'tools/generate_training_knowledge.py',
+            meson.project_source_root() / 'catalog', '@OUTPUT@'],
+)
+
 strict_c_args = [
   '-D_POSIX_C_SOURCE=200809L',
   '-Wconversion',
@@ -45,15 +67,18 @@ trainlog_core_sources = files(
   'src/duration.c',
   'src/id.c',
   'src/timeutil.c',
+  'src/training_context.c',
   'src/usb.c',
   'src/mtp.c',
   'src/reps.c',
   'src/measured_max.c',
   'src/sync.c',
   'src/sync_history.c',
+  'src/timestamp.c',
 )
 trainlog_core_sources += equipment_catalog_generated
 trainlog_core_sources += body_zone_catalog_generated
+trainlog_core_sources += training_knowledge_generated
 
 trainlog_core = static_library(
   'trainlog_core',
@@ -142,6 +167,22 @@ test_body_zones = executable(
 )
 test('body_zones', test_body_zones)
 
+test_training_knowledge = executable(
+  'test_training_knowledge',
+  'tests/test_training_knowledge.c',
+  dependencies: trainlog_core_dep,
+  c_args: strict_c_args,
+)
+test('training_knowledge', test_training_knowledge)
+
+test_training_context = executable(
+  'test_training_context',
+  'tests/test_training_context.c',
+  dependencies: trainlog_core_dep,
+  c_args: strict_c_args,
+)
+test('training_context', test_training_context)
+
 test_custom_equipment = executable(
   'test_custom_equipment',
   'tests/test_custom_equipment.c',
@@ -399,6 +440,12 @@ test(
 
 python3_trainlog_tests = find_program('python3')
 
+test(
+  'timestamp_validation',
+  python3_trainlog_tests,
+  args: [meson.project_source_root() / 'tests/test_timestamp_validation.py'],
+)
+
 test(
   'mobile_import_variable_sets',
   python3_trainlog_tests,

tui/src/database.c

@@ -8,8 +8,10 @@
 #include "trainlog/duration.h"
 #include "trainlog/equipment_catalog.h"
 #include "trainlog/id.h"
+#include "timestamp.h"
 
 #include <math.h>
+#include <limits.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -19,8 +21,38 @@
 
 struct TrainlogDatabase {
     sqlite3 *connection;
+    unsigned int read_snapshot_depth;
 };
 
+TrainlogStatus trainlog_database_read_snapshot_begin(TrainlogDatabase *database)
+{
+    char sql[64];
+    if (database == NULL || database->connection == NULL || database->read_snapshot_depth == UINT_MAX)
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    (void)snprintf(sql, sizeof(sql), "SAVEPOINT trainlog_read_%u;", database->read_snapshot_depth);
+    if (sqlite3_exec(database->connection, sql, NULL, NULL, NULL) != SQLITE_OK)
+        return TRAINLOG_STATUS_DATABASE_ERROR;
+    ++database->read_snapshot_depth;
+    return TRAINLOG_STATUS_OK;
+}
+
+TrainlogStatus trainlog_database_read_snapshot_end(TrainlogDatabase *database, bool commit_snapshot)
+{
+    char sql[160];
+    unsigned int depth;
+    if (database == NULL || database->connection == NULL || database->read_snapshot_depth == 0U)
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    depth = database->read_snapshot_depth - 1U;
+    if (commit_snapshot)
+        (void)snprintf(sql, sizeof(sql), "RELEASE trainlog_read_%u;", depth);
+    else
+        (void)snprintf(sql, sizeof(sql), "ROLLBACK TO trainlog_read_%u; RELEASE trainlog_read_%u;", depth, depth);
+    if (sqlite3_exec(database->connection, sql, NULL, NULL, NULL) != SQLITE_OK)
+        return TRAINLOG_STATUS_DATABASE_ERROR;
+    database->read_snapshot_depth = depth;
+    return TRAINLOG_STATUS_OK;
+}
+
 static TrainlogStatus lookup_exercise_row_id(
     TrainlogDatabase *database,
     const char *exercise_id,
@@ -6102,3 +6134,486 @@ TrainlogStatus trainlog_database_update_body_observation(
         ? TRAINLOG_STATUS_OK
         : TRAINLOG_STATUS_DATABASE_ERROR;
 }
+
+/* TRAINING_KNOWLEDGE_RUNTIME_READ_V1 */
+static bool knowledge_copy_column(sqlite3_stmt *statement, int column, char *output, size_t capacity)
+{
+    const unsigned char *value;
+    size_t length;
+    if (statement == NULL || sqlite3_column_type(statement, column) != SQLITE_TEXT ||
+            output == NULL || capacity == 0U) return false;
+    value = sqlite3_column_text(statement, column);
+    if (value == NULL) return false;
+    length = (size_t)sqlite3_column_bytes(statement, column);
+    if (length >= capacity) return false;
+    (void)memcpy(output, value, length + 1U);
+    return true;
+}
+
+static bool knowledge_numeric_column(sqlite3_stmt *statement, int column)
+{
+    int type = sqlite3_column_type(statement, column);
+    return type == SQLITE_INTEGER || type == SQLITE_FLOAT;
+}
+
+static bool knowledge_bounded_string(const char *value, size_t capacity, size_t *length)
+{
+    const char *end;
+    if (value == NULL || capacity == 0U) return false;
+    end = memchr(value, '\0', capacity);
+    if (end == NULL || end == value) return false;
+    if (length != NULL) *length = (size_t)(end - value);
+    return true;
+}
+
+typedef struct KnowledgeTemporalCandidate {
+    sqlite3_int64 row_id;
+    char *started_at;
+    size_t started_at_length;
+    char *session_id;
+    size_t session_id_length;
+    char *entry_id;
+    size_t entry_id_length;
+    TrainlogTimestampKey timestamp;
+} KnowledgeTemporalCandidate;
+
+static void knowledge_temporal_candidate_release(KnowledgeTemporalCandidate *candidate)
+{
+    if (candidate == NULL) return;
+    free(candidate->started_at);
+    free(candidate->session_id);
+    free(candidate->entry_id);
+    (void)memset(candidate, 0, sizeof(*candidate));
+}
+
+static TrainlogStatus knowledge_copy_dynamic_column(
+    sqlite3_stmt *statement, int column, char **output, size_t *output_length)
+{
+    const unsigned char *source;
+    size_t length;
+    char *copy;
+    if (sqlite3_column_type(statement, column) != SQLITE_TEXT) return TRAINLOG_STATUS_DATABASE_ERROR;
+    source = sqlite3_column_text(statement, column);
+    if (source == NULL) return TRAINLOG_STATUS_DATABASE_ERROR;
+    length = (size_t)sqlite3_column_bytes(statement, column);
+    if (length == 0U || memchr(source, '\0', length) != NULL || length == SIZE_MAX)
+        return TRAINLOG_STATUS_DATABASE_ERROR;
+    copy = malloc(length + 1U);
+    if (copy == NULL) return TRAINLOG_STATUS_SYSTEM_ERROR;
+    (void)memcpy(copy, source, length);
+    copy[length] = '\0';
+    *output = copy;
+    *output_length = length;
+    return TRAINLOG_STATUS_OK;
+}
+
+static int knowledge_bytes_compare(
+    const char *left, size_t left_length, const char *right, size_t right_length)
+{
+    size_t common = left_length < right_length ? left_length : right_length;
+    int result = memcmp(left, right, common);
+    if (result != 0) return result;
+    if (left_length == right_length) return 0;
+    return left_length < right_length ? -1 : 1;
+}
+
+/* INVARIANT: This is the single ordering relation used for selection and the
+ * exclusive cursor. Source timestamp spelling never participates in a tie. */
+static int knowledge_temporal_candidate_compare(
+    const KnowledgeTemporalCandidate *left, const KnowledgeTemporalCandidate *right)
+{
+    int result = trainlog_timestamp_compare(&left->timestamp, &right->timestamp);
+    if (result != 0) return result;
+    result = knowledge_bytes_compare(left->session_id, left->session_id_length,
+        right->session_id, right->session_id_length);
+    if (result != 0) return result;
+    return knowledge_bytes_compare(left->entry_id, left->entry_id_length,
+        right->entry_id, right->entry_id_length);
+}
+
+static TrainlogStatus knowledge_temporal_candidate_read(
+    sqlite3_stmt *statement, KnowledgeTemporalCandidate *output)
+{
+    TrainlogStatus status;
+    (void)memset(output, 0, sizeof(*output));
+    if (sqlite3_column_type(statement, 0) != SQLITE_INTEGER) return TRAINLOG_STATUS_DATABASE_ERROR;
+    status = knowledge_copy_dynamic_column(statement, 1, &output->session_id, &output->session_id_length);
+    if (status == TRAINLOG_STATUS_OK)
+        status = knowledge_copy_dynamic_column(statement, 2, &output->entry_id, &output->entry_id_length);
+    if (status == TRAINLOG_STATUS_OK)
+        status = knowledge_copy_dynamic_column(statement, 3, &output->started_at, &output->started_at_length);
+    if (status != TRAINLOG_STATUS_OK ||
+            !trainlog_timestamp_parse(output->started_at, output->started_at_length, &output->timestamp)) {
+        knowledge_temporal_candidate_release(output);
+        return status == TRAINLOG_STATUS_OK ? TRAINLOG_STATUS_DATABASE_ERROR : status;
+    }
+    output->row_id = sqlite3_column_int64(statement, 0);
+    return TRAINLOG_STATUS_OK;
+}
+
+/* Retain a descending prefix. Capacity is at most page limit + one, so a
+ * sorted bounded array keeps memory independent of history cardinality. */
+static void knowledge_temporal_candidate_insert(
+    KnowledgeTemporalCandidate *items, size_t *count, size_t capacity,
+    KnowledgeTemporalCandidate *candidate)
+{
+    size_t position = 0U;
+    size_t index;
+    while (position < *count &&
+            knowledge_temporal_candidate_compare(candidate, &items[position]) <= 0) ++position;
+    if (position >= capacity) {
+        knowledge_temporal_candidate_release(candidate);
+        return;
+    }
+    if (*count == capacity) knowledge_temporal_candidate_release(&items[capacity - 1U]);
+    else ++*count;
+    for (index = *count - 1U; index > position; --index) items[index] = items[index - 1U];
+    items[position] = *candidate;
+    (void)memset(candidate, 0, sizeof(*candidate));
+}
+
+static bool knowledge_tracking_mode(const char *value, TrainlogTrackingMode *output)
+{
+    if (value == NULL || output == NULL) return false;
+    if (strcmp(value,"reps")==0) {*output=TRAINLOG_TRACKING_REPS;return true;}
+    if (strcmp(value,"duration")==0) {*output=TRAINLOG_TRACKING_DURATION;return true;}
+    return false;
+}
+
+static bool knowledge_recording_mode(const char *value, TrainlogRecordingMode *output)
+{
+    if (value == NULL || output == NULL) return false;
+    if (strcmp(value,"sets")==0) {*output=TRAINLOG_RECORDING_SETS;return true;}
+    if (strcmp(value,"continuous")==0) {*output=TRAINLOG_RECORDING_CONTINUOUS;return true;}
+    return false;
+}
+
+static bool knowledge_load_mode(const char *value, TrainlogLoadMode *output)
+{
+    if (value == NULL || output == NULL) return false;
+    if (strcmp(value,"none")==0) {*output=TRAINLOG_LOAD_NONE;return true;}
+    if (strcmp(value,"external")==0) {*output=TRAINLOG_LOAD_EXTERNAL;return true;}
+    if (strcmp(value,"assistance")==0) {*output=TRAINLOG_LOAD_ASSISTANCE;return true;}
+    return false;
+}
+
+TrainlogStatus trainlog_database_get_exercise_profile(
+    TrainlogDatabase *database, const char *exercise_id, TrainlogExercise *output)
+{
+    sqlite3_stmt *statement = NULL;
+    int rc;
+    if (database == NULL || database->connection == NULL || exercise_id == NULL || exercise_id[0] == '\0' ||
+            output == NULL) return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    (void)memset(output, 0, sizeof(*output));
+    rc = sqlite3_prepare_v2(database->connection,
+        "SELECT exercise_id,name,tracking_mode,recording_mode,data_fields FROM exercises WHERE exercise_id=?1;",
+        -1, &statement, NULL);
+    if (rc == SQLITE_OK) rc = sqlite3_bind_text(statement, 1, exercise_id, -1, SQLITE_TRANSIENT);
+    if (rc != SQLITE_OK) { (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR; }
+    rc = sqlite3_step(statement);
+    if (rc == SQLITE_DONE) { (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_NOT_FOUND; }
+    if (rc != SQLITE_ROW || !knowledge_copy_column(statement, 0, output->exercise_id, sizeof(output->exercise_id)) ||
+            !knowledge_copy_column(statement, 1, output->name, sizeof(output->name)) ||
+            sqlite3_column_type(statement, 4) != SQLITE_INTEGER) {
+        (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+    }
+    if (sqlite3_column_type(statement,2)!=SQLITE_TEXT || sqlite3_column_type(statement,3)!=SQLITE_TEXT ||
+            !knowledge_tracking_mode((const char *)sqlite3_column_text(statement,2),&output->tracking_mode) ||
+            !knowledge_recording_mode((const char *)sqlite3_column_text(statement,3),&output->recording_mode)) {
+        (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+    }
+    if ((sqlite3_column_int64(statement, 4) < 0) ||
+            ((sqlite3_uint64)sqlite3_column_int64(statement, 4) > UINT32_MAX)) {
+        (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+    }
+    output->data_fields = (TrainlogExerciseDataFields)sqlite3_column_int64(statement, 4);
+    return sqlite3_finalize(statement) == SQLITE_OK ? TRAINLOG_STATUS_OK : TRAINLOG_STATUS_DATABASE_ERROR;
+}
+
+TrainlogStatus trainlog_database_list_exercise_occurrences_page(
+    TrainlogDatabase *database, const char *exercise_id, const TrainlogExerciseOccurrenceCursor *after,
+    size_t limit, TrainlogExerciseOccurrence *output, size_t *output_count, bool *output_has_more,
+    TrainlogExerciseOccurrenceCursor *output_next)
+{
+    static const char *const SCAN_SQL =
+        "SELECT se.id,s.session_id,se.entry_id,s.started_at FROM session_exercises se "
+        "JOIN sessions s ON s.id=se.session_row_id JOIN exercises e ON e.id=se.exercise_row_id "
+        "WHERE e.exercise_id=?1;";
+    static const char *const HYDRATE_SQL =
+        "SELECT s.session_id,se.entry_id,e.exercise_id,s.started_at,COALESCE(se.equipment_id,''),"
+        "s.session_type,e.tracking_mode,se.recording_mode,se.data_fields,se.load_mode,"
+        "(SELECT COUNT(*) FROM performed_sets ps WHERE ps.session_exercise_row_id=se.id),"
+        "ca.duration_seconds,ca.speed_kmh,ca.distance_km FROM session_exercises se "
+        "JOIN sessions s ON s.id=se.session_row_id JOIN exercises e ON e.id=se.exercise_row_id "
+        "LEFT JOIN continuous_activity ca ON ca.session_exercise_row_id=se.id WHERE se.id=?1;";
+    sqlite3_stmt *statement = NULL;
+    KnowledgeTemporalCandidate selected[TRAINLOG_OCCURRENCE_PAGE_MAX + 1U] = {{0}};
+    KnowledgeTemporalCandidate cursor = {0};
+    TrainlogExerciseOccurrenceCursor after_value;
+    TrainlogExercise profile;
+    TrainlogStatus status;
+    size_t selected_count = 0U, count = 0U, index;
+    size_t cursor_started_length = 0U, cursor_session_length = 0U, cursor_entry_length = 0U;
+    int rc;
+    bool snapshot = false;
+    if (output_count != NULL) *output_count = 0U;
+    if (output_has_more != NULL) *output_has_more = false;
+    if (database == NULL || database->connection == NULL || exercise_id == NULL || exercise_id[0] == '\0' ||
+            limit == 0U || limit > TRAINLOG_OCCURRENCE_PAGE_MAX || output == NULL || output_count == NULL ||
+            output_has_more == NULL || output_next == NULL ||
+            (after != NULL && (!knowledge_bounded_string(after->started_at, sizeof(after->started_at),
+                    &cursor_started_length) ||
+                !knowledge_bounded_string(after->session_id, sizeof(after->session_id), &cursor_session_length) ||
+                !knowledge_bounded_string(after->entry_id, sizeof(after->entry_id), &cursor_entry_length) ||
+                !trainlog_timestamp_parse(after->started_at, cursor_started_length, &cursor.timestamp))))
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    if (after != NULL) {
+        after_value = *after;
+        cursor.started_at = after_value.started_at; cursor.started_at_length = cursor_started_length;
+        cursor.session_id = after_value.session_id; cursor.session_id_length = cursor_session_length;
+        cursor.entry_id = after_value.entry_id; cursor.entry_id_length = cursor_entry_length;
+        if (!trainlog_timestamp_parse(cursor.started_at, cursor_started_length, &cursor.timestamp))
+            return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    }
+    status = trainlog_database_read_snapshot_begin(database);
+    if (status != TRAINLOG_STATUS_OK) return status;
+    snapshot = true;
+    status = trainlog_database_get_exercise_profile(database, exercise_id, &profile);
+    if (status != TRAINLOG_STATUS_OK) goto done;
+    (void)memset(output_next, 0, sizeof(*output_next));
+    rc = sqlite3_prepare_v2(database->connection, SCAN_SQL, -1, &statement, NULL);
+    if (rc == SQLITE_OK) rc = sqlite3_bind_text(statement, 1, exercise_id, -1, SQLITE_TRANSIENT);
+    if (rc != SQLITE_OK) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+    while ((rc = sqlite3_step(statement)) == SQLITE_ROW) {
+        KnowledgeTemporalCandidate candidate;
+        status = knowledge_temporal_candidate_read(statement, &candidate);
+        if (status != TRAINLOG_STATUS_OK) goto done;
+        if (after == NULL || knowledge_temporal_candidate_compare(&candidate, &cursor) < 0)
+            knowledge_temporal_candidate_insert(selected, &selected_count, limit + 1U, &candidate);
+        knowledge_temporal_candidate_release(&candidate);
+    }
+    if (rc != SQLITE_DONE) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+    rc = sqlite3_finalize(statement); statement = NULL;
+    if (rc != SQLITE_OK) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+    statement = NULL;
+    *output_has_more = selected_count > limit;
+    count = *output_has_more ? limit : selected_count;
+    for (index = 0U; index < count; ++index) {
+        TrainlogExerciseOccurrence *item;
+        sqlite3_int64 fields;
+        sqlite3_int64 sets;
+        rc = sqlite3_prepare_v2(database->connection, HYDRATE_SQL, -1, &statement, NULL);
+        if (rc == SQLITE_OK) rc = sqlite3_bind_int64(statement, 1, selected[index].row_id);
+        if (rc == SQLITE_OK) rc = sqlite3_step(statement);
+        if (rc != SQLITE_ROW) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+        item = &output[index]; (void)memset(item, 0, sizeof(*item));
+        if (!knowledge_copy_column(statement,0,item->session_id,sizeof(item->session_id)) ||
+            !knowledge_copy_column(statement,1,item->entry_id,sizeof(item->entry_id)) ||
+            !knowledge_copy_column(statement,2,item->exercise_id,sizeof(item->exercise_id)) ||
+            !knowledge_copy_column(statement,3,item->started_at,sizeof(item->started_at)) ||
+            !knowledge_copy_column(statement,4,item->equipment_id,sizeof(item->equipment_id)) ||
+            sqlite3_column_type(statement,5)!=SQLITE_TEXT ||
+            !session_type_from_sql((const char *)sqlite3_column_text(statement,5),&item->session_type)) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        if(sqlite3_column_type(statement,6)!=SQLITE_TEXT || sqlite3_column_type(statement,7)!=SQLITE_TEXT ||
+                sqlite3_column_type(statement,9)!=SQLITE_TEXT ||
+                !knowledge_tracking_mode((const char *)sqlite3_column_text(statement,6),&item->tracking_mode) ||
+                !knowledge_recording_mode((const char *)sqlite3_column_text(statement,7),&item->recording_mode) ||
+                !knowledge_load_mode((const char *)sqlite3_column_text(statement,9),&item->load_mode)) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        if (sqlite3_column_type(statement,8)!=SQLITE_INTEGER || sqlite3_column_type(statement,10)!=SQLITE_INTEGER) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        fields=sqlite3_column_int64(statement,8); sets=sqlite3_column_int64(statement,10);
+        if (fields<0 || (sqlite3_uint64)fields>UINT32_MAX || sets<0 || (sqlite3_uint64)sets>SIZE_MAX) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        item->data_fields=(TrainlogExerciseDataFields)fields;
+        item->set_count=(size_t)sets;
+        if (sqlite3_column_type(statement,11)!=SQLITE_NULL) {
+            sqlite3_int64 duration;
+            if (sqlite3_column_type(statement,11)!=SQLITE_INTEGER) {
+                status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+            }
+            duration=sqlite3_column_int64(statement,11);
+            if (duration<=0 || duration>INT_MAX) {
+                status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+            }
+            item->continuous_duration_seconds=(int)duration;
+        }
+        item->continuous_has_speed=sqlite3_column_type(statement,12)!=SQLITE_NULL;
+        if (item->continuous_has_speed && !knowledge_numeric_column(statement,12)) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        item->continuous_speed_kmh=sqlite3_column_double(statement,12);
+        item->continuous_has_distance=sqlite3_column_type(statement,13)!=SQLITE_NULL;
+        if (item->continuous_has_distance && !knowledge_numeric_column(statement,13)) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        item->continuous_distance_km=sqlite3_column_double(statement,13);
+        if ((item->continuous_has_speed && !isfinite(item->continuous_speed_kmh)) ||
+                (item->continuous_has_distance && !isfinite(item->continuous_distance_km))) {
+            status = TRAINLOG_STATUS_DATABASE_ERROR; goto done;
+        }
+        rc = sqlite3_step(statement);
+        if (rc != SQLITE_DONE) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+        rc = sqlite3_finalize(statement); statement = NULL;
+        if (rc != SQLITE_OK) { status = TRAINLOG_STATUS_DATABASE_ERROR; goto done; }
+    }
+    *output_count=count;
+    if (count > 0U) {
+        (void)snprintf(output_next->started_at,sizeof(output_next->started_at),"%s",output[count-1U].started_at);
+        (void)snprintf(output_next->session_id,sizeof(output_next->session_id),"%s",output[count-1U].session_id);
+        (void)snprintf(output_next->entry_id,sizeof(output_next->entry_id),"%s",output[count-1U].entry_id);
+    }
+    status = TRAINLOG_STATUS_OK;
+done:
+    (void)sqlite3_finalize(statement);
+    for (index = 0U; index < selected_count; ++index)
+        knowledge_temporal_candidate_release(&selected[index]);
+    if (snapshot) {
+        TrainlogStatus end_status = trainlog_database_read_snapshot_end(database, status == TRAINLOG_STATUS_OK);
+        if (status == TRAINLOG_STATUS_OK) status = end_status;
+    }
+    return status;
+}
+
+TrainlogStatus trainlog_database_list_occurrence_sets_page(
+    TrainlogDatabase *database, const char *entry_id, int after_position, size_t limit,
+    TrainlogOccurrenceSet *output, size_t *output_count, bool *output_has_more, int *output_next_position)
+{
+    sqlite3_stmt *statement = NULL;
+    sqlite3_stmt *identity = NULL;
+    size_t count=0U;
+    int rc;
+    if (output_count != NULL) *output_count=0U;
+    if (output_has_more != NULL) *output_has_more=false;
+    if (database==NULL || database->connection==NULL || entry_id==NULL || entry_id[0]=='\0' ||
+            after_position < -1 || limit==0U || limit>TRAINLOG_OCCURRENCE_SET_PAGE_MAX || output==NULL ||
+            output_count==NULL || output_has_more==NULL || output_next_position==NULL)
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    *output_next_position=after_position;
+    rc=sqlite3_prepare_v2(database->connection,"SELECT 1 FROM session_exercises WHERE entry_id=?1;",-1,&identity,NULL);
+    if(rc==SQLITE_OK)rc=sqlite3_bind_text(identity,1,entry_id,-1,SQLITE_TRANSIENT);
+    if(rc!=SQLITE_OK){(void)sqlite3_finalize(identity);return TRAINLOG_STATUS_DATABASE_ERROR;}
+    rc=sqlite3_step(identity);
+    if(rc==SQLITE_DONE){(void)sqlite3_finalize(identity);return TRAINLOG_STATUS_NOT_FOUND;}
+    if(rc!=SQLITE_ROW || sqlite3_finalize(identity)!=SQLITE_OK)return TRAINLOG_STATUS_DATABASE_ERROR;
+    rc=sqlite3_prepare_v2(database->connection,
+        "SELECT ps.position,ps.reps,ps.duration_seconds,ps.weight_kg FROM performed_sets ps "
+        "JOIN session_exercises se ON se.id=ps.session_exercise_row_id WHERE se.entry_id=?1 AND ps.position>?2 "
+        "ORDER BY ps.position ASC LIMIT ?3;",-1,&statement,NULL);
+    if(rc==SQLITE_OK) rc=sqlite3_bind_text(statement,1,entry_id,-1,SQLITE_TRANSIENT);
+    if(rc==SQLITE_OK) rc=sqlite3_bind_int(statement,2,after_position);
+    if(rc==SQLITE_OK) rc=sqlite3_bind_int64(statement,3,(sqlite3_int64)(limit+1U));
+    if(rc!=SQLITE_OK){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+    while((rc=sqlite3_step(statement))==SQLITE_ROW){ TrainlogOccurrenceSet *item; sqlite3_int64 position;
+        if(count==limit){*output_has_more=true;break;} item=&output[count];(void)memset(item,0,sizeof(*item));
+        if(sqlite3_column_type(statement,0)!=SQLITE_INTEGER){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+        position=sqlite3_column_int64(statement,0); if(position<0 || position>INT_MAX){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+        item->position=(size_t)position; item->has_reps=sqlite3_column_type(statement,1)!=SQLITE_NULL;
+        item->has_duration=sqlite3_column_type(statement,2)!=SQLITE_NULL;
+        if (item->has_reps) {
+            sqlite3_int64 reps;
+            if (sqlite3_column_type(statement,1)!=SQLITE_INTEGER) {
+                (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+            }
+            reps=sqlite3_column_int64(statement,1);
+            if (reps<0 || reps>INT_MAX) {
+                (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+            }
+            item->reps=(int)reps;
+        }
+        if (item->has_duration) {
+            sqlite3_int64 duration;
+            if (sqlite3_column_type(statement,2)!=SQLITE_INTEGER) {
+                (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+            }
+            duration=sqlite3_column_int64(statement,2);
+            if (duration<=0 || duration>INT_MAX) {
+                (void)sqlite3_finalize(statement); return TRAINLOG_STATUS_DATABASE_ERROR;
+            }
+            item->duration_seconds=(int)duration;
+        }
+        item->has_weight=sqlite3_column_type(statement,3)!=SQLITE_NULL;
+        if(item->has_weight && !knowledge_numeric_column(statement,3)){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+        item->weight_kg=sqlite3_column_double(statement,3);
+        if(item->has_weight && !isfinite(item->weight_kg)){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+        *output_next_position=(int)position; ++count;
+    }
+    if(rc!=SQLITE_DONE && rc!=SQLITE_ROW){(void)sqlite3_finalize(statement);return TRAINLOG_STATUS_DATABASE_ERROR;}
+    if(sqlite3_finalize(statement)!=SQLITE_OK)return TRAINLOG_STATUS_DATABASE_ERROR;
+    *output_count=count; return TRAINLOG_STATUS_OK;
+}
+
+TrainlogStatus trainlog_database_latest_explicit_max_context(
+    TrainlogDatabase *database, const char *exercise_id, TrainlogLatestExplicitMax *output)
+{
+    static const char *const SCAN_SQL =
+        "SELECT se.id,s.session_id,se.entry_id,s.started_at FROM max_results mr "
+        "JOIN session_exercises se ON se.id=mr.session_exercise_row_id "
+        "JOIN sessions s ON s.id=se.session_row_id JOIN exercises e ON e.id=se.exercise_row_id "
+        "WHERE e.exercise_id=?1 AND s.session_type='max_test';";
+    static const char *const HYDRATE_SQL =
+        "SELECT s.session_id,se.entry_id,s.started_at,COALESCE(se.equipment_id,''),se.load_mode,mr.max_weight_kg "
+        "FROM max_results mr JOIN session_exercises se ON se.id=mr.session_exercise_row_id "
+        "JOIN sessions s ON s.id=se.session_row_id WHERE se.id=?1;";
+    sqlite3_stmt *statement=NULL;
+    KnowledgeTemporalCandidate selected[1] = {{0}};
+    size_t selected_count = 0U;
+    int rc;
+    TrainlogExercise profile;
+    TrainlogStatus status;
+    bool snapshot = false;
+    if(database==NULL || database->connection==NULL || exercise_id==NULL || exercise_id[0]=='\0' || output==NULL)
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    (void)memset(output,0,sizeof(*output));
+    status = trainlog_database_read_snapshot_begin(database);
+    if (status != TRAINLOG_STATUS_OK) return status;
+    snapshot = true;
+    status=trainlog_database_get_exercise_profile(database,exercise_id,&profile);
+    if(status!=TRAINLOG_STATUS_OK)goto done;
+    rc=sqlite3_prepare_v2(database->connection,SCAN_SQL,-1,&statement,NULL);
+    if(rc==SQLITE_OK)rc=sqlite3_bind_text(statement,1,exercise_id,-1,SQLITE_TRANSIENT);
+    if(rc!=SQLITE_OK){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    while((rc=sqlite3_step(statement))==SQLITE_ROW){
+        KnowledgeTemporalCandidate candidate;
+        status=knowledge_temporal_candidate_read(statement,&candidate);
+        if(status!=TRAINLOG_STATUS_OK)goto done;
+        knowledge_temporal_candidate_insert(selected,&selected_count,1U,&candidate);
+        knowledge_temporal_candidate_release(&candidate);
+    }
+    if(rc!=SQLITE_DONE){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    rc=sqlite3_finalize(statement); statement=NULL;
+    if(rc!=SQLITE_OK){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    if(selected_count==0U){status=TRAINLOG_STATUS_OK;goto done;}
+    rc=sqlite3_prepare_v2(database->connection,HYDRATE_SQL,-1,&statement,NULL);
+    if(rc==SQLITE_OK)rc=sqlite3_bind_int64(statement,1,selected[0].row_id);
+    if(rc==SQLITE_OK)rc=sqlite3_step(statement);
+    if(rc!=SQLITE_ROW || !knowledge_copy_column(statement,0,output->session_id,sizeof(output->session_id)) ||
+        !knowledge_copy_column(statement,1,output->entry_id,sizeof(output->entry_id)) ||
+        !knowledge_copy_column(statement,2,output->started_at,sizeof(output->started_at)) ||
+        !knowledge_copy_column(statement,3,output->equipment_id,sizeof(output->equipment_id))){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    if(sqlite3_column_type(statement,4)!=SQLITE_TEXT ||
+            !knowledge_load_mode((const char *)sqlite3_column_text(statement,4),&output->load_mode)){
+        status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;
+    }
+    if(!knowledge_numeric_column(statement,5)){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    output->max_weight_kg=sqlite3_column_double(statement,5);
+    if(!isfinite(output->max_weight_kg) || output->max_weight_kg<=0.0){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    output->found=true;
+    rc=sqlite3_step(statement);
+    if(rc!=SQLITE_DONE){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+    rc=sqlite3_finalize(statement); statement=NULL;
+    status=rc==SQLITE_OK?TRAINLOG_STATUS_OK:TRAINLOG_STATUS_DATABASE_ERROR;
+done:
+    (void)sqlite3_finalize(statement);
+    knowledge_temporal_candidate_release(&selected[0]);
+    if(snapshot){
+        TrainlogStatus end_status=trainlog_database_read_snapshot_end(database,status==TRAINLOG_STATUS_OK);
+        if(status==TRAINLOG_STATUS_OK)status=end_status;
+    }
+    return status;
+}

tui/src/timestamp.c

@@ -0,0 +1,108 @@
+#include "timestamp.h"
+
+#include <limits.h>
+
+static bool parse_digits(const char *value, size_t count, int *output)
+{
+    size_t index;
+    int result = 0;
+    for (index = 0U; index < count; ++index) {
+        if (value[index] < '0' || value[index] > '9') return false;
+        result = result * 10 + (value[index] - '0');
+    }
+    *output = result;
+    return true;
+}
+
+static bool leap_year(int year)
+{
+    return year % 4 == 0 && (year % 100 != 0 || year % 400 == 0);
+}
+
+static int days_in_month(int year, int month)
+{
+    static const int DAYS[] = {31,28,31,30,31,30,31,31,30,31,30,31};
+    return month == 2 && leap_year(year) ? 29 : DAYS[month - 1];
+}
+
+/* Proleptic Gregorian day number relative to 0001-01-01. */
+static int64_t day_number(int year, int month, int day)
+{
+    int prior_year = year - 1;
+    int64_t days = (int64_t)prior_year * 365 + prior_year / 4 -
+        prior_year / 100 + prior_year / 400;
+    int current_month;
+    for (current_month = 1; current_month < month; ++current_month)
+        days += days_in_month(year, current_month);
+    return days + day - 1;
+}
+
+bool trainlog_timestamp_parse(
+    const char *value, size_t length, TrainlogTimestampKey *output)
+{
+    size_t zone;
+    int year, month, day, hour, minute, second = 0;
+    int offset_hour = 0, offset_minute = 0, offset_sign = 0;
+    int64_t local_second;
+
+    if (value == NULL || output == NULL || length < 17U ||
+            !parse_digits(value, 4U, &year) || value[4] != '-' ||
+            !parse_digits(value + 5, 2U, &month) || value[7] != '-' ||
+            !parse_digits(value + 8, 2U, &day) ||
+            (value[10] != 'T' && value[10] != 't') ||
+            !parse_digits(value + 11, 2U, &hour) || value[13] != ':' ||
+            !parse_digits(value + 14, 2U, &minute)) return false;
+    if (year < 1 || month < 1 || month > 12 || day < 1 ||
+            day > days_in_month(year, month) || hour > 23 || minute > 59)
+        return false;
+
+    zone = 16U;
+    output->fraction = NULL;
+    output->fraction_length = 0U;
+    if (zone < length && value[zone] == ':') {
+        if (zone + 3U > length || !parse_digits(value + zone + 1U, 2U, &second) || second > 59)
+            return false;
+        zone += 3U;
+        if (zone < length && value[zone] == '.') {
+            size_t fraction_start = ++zone;
+            while (zone < length && value[zone] >= '0' && value[zone] <= '9') ++zone;
+            if (zone == fraction_start) return false;
+            output->fraction = value + fraction_start;
+            output->fraction_length = zone - fraction_start;
+        }
+    }
+
+    if (zone + 1U == length && (value[zone] == 'Z' || value[zone] == 'z')) {
+        offset_sign = 0;
+    } else {
+        if (zone + 6U != length || (value[zone] != '+' && value[zone] != '-') ||
+                value[zone + 3U] != ':' ||
+                !parse_digits(value + zone + 1U, 2U, &offset_hour) ||
+                !parse_digits(value + zone + 4U, 2U, &offset_minute) ||
+                offset_hour > 23 || offset_minute > 59) return false;
+        offset_sign = value[zone] == '+' ? 1 : -1;
+    }
+
+    local_second = day_number(year, month, day) * INT64_C(86400) +
+        (int64_t)hour * INT64_C(3600) + (int64_t)minute * INT64_C(60) + second;
+    output->utc_second = local_second - offset_sign *
+        ((int64_t)offset_hour * INT64_C(3600) + (int64_t)offset_minute * INT64_C(60));
+    return true;
+}
+
+int trainlog_timestamp_compare(
+    const TrainlogTimestampKey *left, const TrainlogTimestampKey *right)
+{
+    size_t index;
+    size_t count;
+    if (left->utc_second != right->utc_second)
+        return left->utc_second < right->utc_second ? -1 : 1;
+    count = left->fraction_length > right->fraction_length
+        ? left->fraction_length : right->fraction_length;
+    for (index = 0U; index < count; ++index) {
+        char left_digit = index < left->fraction_length ? left->fraction[index] : '0';
+        char right_digit = index < right->fraction_length ? right->fraction[index] : '0';
+        if (left_digit != right_digit) return left_digit < right_digit ? -1 : 1;
+    }
+    return 0;
+}

tui/src/timestamp.h

@@ -0,0 +1,29 @@
+#ifndef TRAINLOG_INTERNAL_TIMESTAMP_H
+#define TRAINLOG_INTERNAL_TIMESTAMP_H
+
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+typedef struct TrainlogTimestampKey {
+    int64_t utc_second;
+    const char *fraction;
+    size_t fraction_length;
+} TrainlogTimestampKey;
+
+/* CONTRACT: Parses the frozen Trainlog timestamp grammar from exactly length
+ * bytes. The key borrows fraction storage from value and remains valid only
+ * while value does. */
+bool trainlog_timestamp_parse(
+    const char *value,
+    size_t length,
+    TrainlogTimestampKey *output
+);
+
+/* Returns less than, zero, or greater than zero by exact represented instant. */
+int trainlog_timestamp_compare(
+    const TrainlogTimestampKey *left,
+    const TrainlogTimestampKey *right
+);
+
+#endif

tui/src/training_context.c

@@ -0,0 +1,118 @@
+#include "trainlog/training_context.h"
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+static size_t id_list_count(const char *list)
+{
+    size_t count = 0U;
+    const char *at = list;
+    if (at == NULL || *at == '\0') return 0U;
+    count = 1U;
+    while (*at != '\0') { if (*at == '\n') ++count; ++at; }
+    return count;
+}
+
+static const TrainlogEquipmentKnowledge *equipment_from_list(const char **at)
+{
+    const char *end;
+    char id[TRAINLOG_ID_MAX + 1U];
+    size_t length;
+    if (at == NULL || *at == NULL || **at == '\0') return NULL;
+    end = strchr(*at, '\n');
+    length = end == NULL ? strlen(*at) : (size_t)(end - *at);
+    if (length == 0U || length >= sizeof(id)) return NULL;
+    (void)memcpy(id, *at, length); id[length] = '\0';
+    *at = end == NULL ? *at + length : end + 1;
+    return trainlog_equipment_knowledge_lookup(id);
+}
+
+void trainlog_training_exercise_context_release(TrainlogTrainingExerciseContext *context)
+{
+    if (context == NULL) return;
+    free(context->persisted_zones);
+    free(context->compatible_equipment);
+    free(context->occurrences);
+    free(context->sets);
+    (void)memset(context, 0, sizeof(*context));
+}
+
+TrainlogStatus trainlog_training_exercise_context_load(
+    TrainlogDatabase *database, const char *exercise_id, size_t occurrence_limit,
+    size_t set_preview_limit, TrainlogTrainingExerciseContext *output)
+{
+    TrainlogTrainingExerciseContext result = {0};
+    TrainlogStatus status;
+    const char *equipment_at;
+    size_t index;
+    size_t zone_count = 0U;
+    bool snapshot = false;
+    if (database == NULL || exercise_id == NULL || exercise_id[0] == '\0' || output == NULL ||
+            occurrence_limit == 0U || occurrence_limit > TRAINLOG_OCCURRENCE_PAGE_MAX ||
+            set_preview_limit == 0U || set_preview_limit > TRAINLOG_OCCURRENCE_SET_PAGE_MAX)
+        return TRAINLOG_STATUS_INVALID_ARGUMENT;
+    (void)memset(output, 0, sizeof(*output));
+    status = trainlog_database_read_snapshot_begin(database);
+    if (status != TRAINLOG_STATUS_OK) return status;
+    snapshot = true;
+    status = trainlog_database_get_exercise_profile(database, exercise_id, &result.exercise);
+    if (status != TRAINLOG_STATUS_OK) goto done;
+    status = trainlog_database_list_exercise_body_zones(database, exercise_id, NULL, 0U, &zone_count);
+    if (status != TRAINLOG_STATUS_OK && !(status == TRAINLOG_STATUS_INVALID_ARGUMENT && zone_count > 0U)) goto done;
+    if (zone_count > SIZE_MAX / sizeof(*result.persisted_zones)) { status=TRAINLOG_STATUS_INVALID_ARGUMENT; goto done; }
+    if (zone_count > 0U) {
+        result.persisted_zones = calloc(zone_count, sizeof(*result.persisted_zones));
+        if (result.persisted_zones == NULL) { status=TRAINLOG_STATUS_SYSTEM_ERROR; goto done; }
+        status=trainlog_database_list_exercise_body_zones(database,exercise_id,result.persisted_zones,zone_count,&result.persisted_zone_count);
+        if(status!=TRAINLOG_STATUS_OK)goto done;
+    }
+    result.knowledge=trainlog_exercise_knowledge_lookup(exercise_id);
+    if(result.knowledge!=NULL){
+        result.compatible_equipment_count=id_list_count(result.knowledge->equipment_ids);
+        if(result.compatible_equipment_count>SIZE_MAX/sizeof(*result.compatible_equipment)){status=TRAINLOG_STATUS_INVALID_ARGUMENT;goto done;}
+        if(result.compatible_equipment_count>0U){
+            result.compatible_equipment=calloc(result.compatible_equipment_count,sizeof(*result.compatible_equipment));
+            if(result.compatible_equipment==NULL){status=TRAINLOG_STATUS_SYSTEM_ERROR;goto done;}
+            equipment_at=result.knowledge->equipment_ids;
+            for(index=0U;index<result.compatible_equipment_count;++index){
+                result.compatible_equipment[index]=equipment_from_list(&equipment_at);
+                if(result.compatible_equipment[index]==NULL){status=TRAINLOG_STATUS_DATABASE_ERROR;goto done;}
+            }
+        }
+    }
+    status=trainlog_database_latest_explicit_max_context(database,exercise_id,&result.latest_max);
+    if(status!=TRAINLOG_STATUS_OK)goto done;
+    result.occurrences=calloc(occurrence_limit,sizeof(*result.occurrences));
+    if(result.occurrences==NULL){status=TRAINLOG_STATUS_SYSTEM_ERROR;goto done;}
+    {
+        TrainlogExerciseOccurrence *raw=calloc(occurrence_limit,sizeof(*raw));
+        if(raw==NULL){status=TRAINLOG_STATUS_SYSTEM_ERROR;goto done;}
+        status=trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,occurrence_limit,raw,
+            &result.occurrence_count,&result.occurrences_have_more,&result.next_occurrence);
+        if(status==TRAINLOG_STATUS_OK){for(index=0U;index<result.occurrence_count;++index)result.occurrences[index].occurrence=raw[index];}
+        free(raw); if(status!=TRAINLOG_STATUS_OK)goto done;
+    }
+    if(result.occurrence_count>0U && set_preview_limit>SIZE_MAX/result.occurrence_count){status=TRAINLOG_STATUS_INVALID_ARGUMENT;goto done;}
+    result.sets=calloc(result.occurrence_count*set_preview_limit,sizeof(*result.sets));
+    if(result.occurrence_count>0U && result.sets==NULL){status=TRAINLOG_STATUS_SYSTEM_ERROR;goto done;}
+    for(index=0U;index<result.occurrence_count;++index){
+        TrainlogTrainingOccurrenceView *view=&result.occurrences[index]; size_t count=0U;
+        view->set_offset=result.set_count; view->next_set_position=-1;
+        if(view->occurrence.recording_mode==TRAINLOG_RECORDING_CONTINUOUS)continue;
+        status=trainlog_database_list_occurrence_sets_page(database,view->occurrence.entry_id,-1,set_preview_limit,
+            result.sets+result.set_count,&count,&view->sets_have_more,&view->next_set_position);
+        if (status != TRAINLOG_STATUS_OK) {
+            goto done;
+        }
+        view->set_count = count;
+        result.set_count += count;
+    }
+done:
+    if (snapshot) {
+        TrainlogStatus end_status=trainlog_database_read_snapshot_end(database,status==TRAINLOG_STATUS_OK);
+        if(status==TRAINLOG_STATUS_OK)status=end_status;
+    }
+    if(status!=TRAINLOG_STATUS_OK){trainlog_training_exercise_context_release(&result);return status;}
+    *output=result; return TRAINLOG_STATUS_OK;
+}

tui/src/tui.c

@@ -20,6 +20,7 @@
 #include <wchar.h>
 #include <unistd.h>
 
+#include <utf8proc.h>
 #include <uuid/uuid.h>
 
 #include "trainlog/terminal.h"
@@ -39,6 +40,7 @@
 #include "trainlog/reps.h"
 #include "trainlog/theme.h"
 #include "trainlog/timeutil.h"
+#include "trainlog/training_knowledge.h"
 #include "trainlog/usb.h"
 
 #define MAX_EXERCISES 128U
@@ -2380,6 +2382,180 @@ static bool edit_exercise_body_zones(
     return status == TRAINLOG_STATUS_OK;
 }
 
+/* CONTRACT: knowledge text is read-only catalogue data.  The screen stores
+ * display lines locally so navigation never changes the scientific record. */
+#define KNOWLEDGE_LINES_MAX 128U
+#define KNOWLEDGE_LINE_MAX 256U
+
+static void knowledge_add_wrapped(char lines[][KNOWLEDGE_LINE_MAX], size_t *count,
+                                  const char *text, int width)
+{
+    const char *at = text == NULL ? "" : text;
+    size_t used = 0U;
+    int cells = 0;
+
+    /* WHY: terminal columns count display cells, while French labels are UTF-8;
+     * utf8proc prevents a wrap from splitting an accented character or from
+     * writing past the right border for a wide code point. */
+    while (*at != '\0' && *count < KNOWLEDGE_LINES_MAX) {
+        utf8proc_int32_t codepoint;
+        utf8proc_ssize_t bytes = utf8proc_iterate((const utf8proc_uint8_t *)at,
+            -1, &codepoint);
+        int codepoint_cells;
+        if (bytes <= 0) { codepoint = (unsigned char)*at; bytes = 1; }
+        codepoint_cells = utf8proc_charwidth(codepoint);
+        if (codepoint_cells < 0) codepoint_cells = 1;
+        if (cells > 0 && cells + codepoint_cells > width) {
+            lines[*count][used] = '\0';
+            ++*count;
+            used = 0U;
+            cells = 0;
+            continue;
+        }
+        if (used + (size_t)bytes >= KNOWLEDGE_LINE_MAX - 1U) break;
+        (void)memcpy(lines[*count] + used, at, (size_t)bytes);
+        used += (size_t)bytes;
+        cells += codepoint_cells;
+        at += bytes;
+    }
+    /* INVARIANT: every entry is NUL-terminated before rendering, and the fixed
+     * line capacity bounds catalogue display independently of terminal size. */
+    if (*count < KNOWLEDGE_LINES_MAX) {
+        lines[*count][used] = '\0';
+        ++*count;
+    }
+}
+
+static void knowledge_add_ids(char lines[][KNOWLEDGE_LINE_MAX], size_t *count,
+                              const char *heading, const char *ids, bool muscles,
+                              int width)
+{
+    const char *at = ids;
+    knowledge_add_wrapped(lines, count, heading, width);
+    while (at != NULL && *at != '\0' && *count < KNOWLEDGE_LINES_MAX) {
+        const char *end = strchr(at, '\n');
+        size_t length = end == NULL ? strlen(at) : (size_t)(end - at);
+        char id[96];
+        const char *label = NULL;
+        if (length >= sizeof(id)) break;
+        (void)memcpy(id, at, length);
+        id[length] = '\0';
+        if (muscles) {
+            const TrainlogKnowledgeMuscle *item = trainlog_knowledge_muscle_lookup(id);
+            if (item != NULL) label = item->display_name_fr;
+        } else {
+            const TrainlogKnowledgeMovementPattern *item =
+                trainlog_knowledge_movement_pattern_lookup(id);
+            if (item != NULL) label = item->display_name_fr;
+        }
+        knowledge_add_wrapped(lines, count, label == NULL ? id : label, width);
+        if (end == NULL) break;
+        at = end + 1;
+    }
+}
+
+static void knowledge_add_plain_ids(char lines[][KNOWLEDGE_LINE_MAX], size_t *count,
+                                    const char *heading, const char *ids, int width)
+{
+    const char *at = ids;
+    knowledge_add_wrapped(lines, count, heading, width);
+    while (at != NULL && *at != '\0' && *count < KNOWLEDGE_LINES_MAX) {
+        const char *end = strchr(at, '\n');
+        size_t length = end == NULL ? strlen(at) : (size_t)(end - at);
+        char id[96];
+        if (length >= sizeof(id)) break;
+        (void)memcpy(id, at, length);
+        id[length] = '\0';
+        knowledge_add_wrapped(lines, count, id, width);
+        if (end == NULL) break;
+        at = end + 1;
+    }
+}
+
+static void knowledge_add_zones(char lines[][KNOWLEDGE_LINE_MAX], size_t *count,
+                                const TrainlogKnowledgeInterpretation *value, int width)
+{
+    const TrainlogBodyZone *zone;
+    knowledge_add_wrapped(lines, count, "Zones scientifiques :", width);
+    zone = trainlog_body_zone_catalog_lookup(value->primary_zone_id);
+    knowledge_add_wrapped(lines, count,
+        zone == NULL ? value->primary_zone_id : zone->display_name, width);
+    knowledge_add_plain_ids(lines, count, "Zones secondaires :",
+        value->secondary_zone_ids, width);
+}
+
+static void screen_exercise_knowledge(const TrainlogExercise *exercise)
+{
+    const TrainlogExerciseKnowledge *record;
+    const TrainlogKnowledgeInterpretation *value;
+    const char *label;
+    char lines[KNOWLEDGE_LINES_MAX][KNOWLEDGE_LINE_MAX];
+    size_t line_count;
+    size_t scroll = 0U;
+    int key;
+    if (exercise == NULL) return;
+    record = trainlog_exercise_knowledge_lookup(exercise->exercise_id);
+    value = record == NULL ? NULL : record->interpretation;
+    label = "Connaissances validées";
+    if (value == NULL && record != NULL && record->conditional_interpretation != NULL) {
+        value = record->conditional_interpretation;
+        label = "Interprétation conditionnelle — à confirmer";
+    }
+    for (;;) {
+        int rows = trainlog_terminal_rows(tui_terminal);
+        int columns = trainlog_terminal_columns(tui_terminal);
+        int viewport_rows;
+        int width;
+        size_t index;
+        if (rows < 20 || columns < 72) {
+            draw_shell("TRAINLOG — Connaissances exercice", "b/Échap retour");
+            trainlog_terminal_printf(tui_terminal, 3, 4, "Terminal trop petit — minimum 72x20.");
+            trainlog_terminal_render(tui_terminal);
+            key = trainlog_terminal_get_key(tui_terminal);
+            if (key == 27 || key == 'b' || key == 'B' || key == 'k' || key == 'K') return;
+            continue;
+        }
+        width = columns - 8;
+        viewport_rows = rows - 6;
+        line_count = 0U;
+        knowledge_add_wrapped(lines, &line_count, exercise->name, width);
+        if (record == NULL || value == NULL) {
+            knowledge_add_wrapped(lines, &line_count,
+                record == NULL ? "Aucune fiche scientifique pour cet identifiant."
+                               : "Interprétation scientifique non résolue.", width);
+        } else {
+            knowledge_add_wrapped(lines, &line_count, label, width);
+            knowledge_add_ids(lines, &line_count, "Mouvement :", value->pattern_ids, false, width);
+            knowledge_add_wrapped(lines, &line_count, "Confiance :", width);
+            knowledge_add_wrapped(lines, &line_count, value->confidence, width);
+            knowledge_add_ids(lines, &line_count, "Muscles principaux :",
+                value->primary_muscle_ids, true, width);
+            knowledge_add_ids(lines, &line_count, "Secondaires :",
+                value->secondary_muscle_ids, true, width);
+            knowledge_add_ids(lines, &line_count, "Stabilisateurs :",
+                value->stabilizer_muscle_ids, true, width);
+            knowledge_add_zones(lines, &line_count, value, width);
+            knowledge_add_plain_ids(lines, &line_count, "Sources :", value->source_refs, width);
+        }
+        if (scroll >= line_count) scroll = line_count == 0U ? 0U : line_count - 1U;
+        draw_shell("TRAINLOG — Connaissances exercice",
+            "↑↓ défiler  Pg↑/Pg↓ page  b/Échap/k retour");
+        for (index = 0U; index < (size_t)viewport_rows && scroll + index < line_count; ++index)
+            trainlog_terminal_printf(tui_terminal, 3 + (int)index, 4, "%s", lines[scroll + index]);
+        trainlog_terminal_render(tui_terminal);
+        key = trainlog_terminal_get_key(tui_terminal);
+        if (key == 27 || key == 'b' || key == 'B' || key == 'k' || key == 'K') return;
+        if (key == TRAINLOG_KEY_UP && scroll > 0U) --scroll;
+        else if (key == TRAINLOG_KEY_DOWN && scroll + (size_t)viewport_rows < line_count) ++scroll;
+        else if (key == TRAINLOG_KEY_PAGE_UP) {
+            scroll = scroll > (size_t)viewport_rows ? scroll - (size_t)viewport_rows : 0U;
+        } else if (key == TRAINLOG_KEY_PAGE_DOWN && scroll + (size_t)viewport_rows < line_count) {
+            size_t maximum = line_count - (size_t)viewport_rows;
+            scroll = scroll + (size_t)viewport_rows < maximum ? scroll + (size_t)viewport_rows : maximum;
+        }
+    }
+}
+
 static void screen_exercise_detail(
     TrainlogDatabase *database,
     TrainlogExercise *exercise
@@ -2418,7 +2594,7 @@ static void screen_exercise_detail(
         int key;
         if (total > 0U && selected >= total) selected = total - 1U;
         draw_shell("TRAINLOG — Fiche exercice",
-            "↑↓ équipement  Entrée fiche  e modifier  p performance  m max mesuré  b/Échap retour");
+            "↑↓ équipement  Entrée fiche  k connaissances  p performance  m max  b/Échap retour");
         trainlog_terminal_printf(tui_terminal, 3, 4, "%s", exercise->name);
         trainlog_terminal_printf(tui_terminal, 4, 4, "Identifiant : %s · suivi : %s",
             exercise->exercise_id,
@@ -2479,6 +2655,7 @@ static void screen_exercise_detail(
                 ? &explicit_items[selected] : &historic_items[selected - explicit_count]);
         else if (key == 'p' || key == 'P') screen_exercise_performance(database, exercise);
         else if (key == 'm' || key == 'M') screen_exercise_measured_max(database, exercise);
+        else if (key == 'k' || key == 'K') screen_exercise_knowledge(exercise);
         else if (key == 'e' || key == 'E') (void)edit_exercise_body_zones(database, exercise);
     }
 }

tui/tests/test_training_context.c

@@ -0,0 +1,362 @@
+#include "trainlog/training_context.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+
+#include <sqlite3.h>
+
+#define CHECK(test) do { if (!(test)) { fprintf(stderr, "CHECK failed: %s:%d: %s\n", \
+    __FILE__, __LINE__, #test); return 1; } } while (0)
+
+static void base_occurrence(TrainlogSessionExerciseInput *value, const char *entry_id,
+    const char *exercise_id, const char *equipment_id, TrainlogLoadMode mode)
+{
+    (void)memset(value, 0, sizeof(*value));
+    (void)snprintf(value->entry_id, sizeof(value->entry_id), "%s", entry_id);
+    (void)snprintf(value->exercise_id, sizeof(value->exercise_id), "%s", exercise_id);
+    (void)snprintf(value->equipment_id, sizeof(value->equipment_id), "%s", equipment_id);
+    value->recording_mode = TRAINLOG_RECORDING_SETS;
+    value->load_mode = mode;
+    value->rest_seconds = 60;
+    value->target_sets = 2;
+    value->target_reps = 10;
+    value->target_has_weight = mode != TRAINLOG_LOAD_NONE;
+    value->target_weight_kg = 20.0;
+}
+
+static int insert_session(TrainlogDatabase *database, const char *session_id, const char *started_at,
+    TrainlogSessionType type, TrainlogSessionExerciseInput *items, size_t count)
+{
+    TrainlogSessionInput session = {0};
+    (void)snprintf(session.session_id, sizeof(session.session_id), "%s", session_id);
+    (void)snprintf(session.started_at, sizeof(session.started_at), "%s", started_at);
+    (void)snprintf(session.ended_at, sizeof(session.ended_at), "%s", started_at);
+    session.session_type = type;
+    session.exercises = items;
+    session.exercise_count = count;
+    return trainlog_database_insert_session(database, &session) == TRAINLOG_STATUS_OK;
+}
+
+int main(void)
+{
+    const char *exercise_id = "ex_b432623f-bfe9-4daf-a653-60ec7fdffbde";
+    const char *zones[] = {"glutes"};
+    TrainlogDatabase *database = NULL;
+    TrainlogSessionExerciseInput occurrences[2];
+    TrainlogSessionExerciseInput max_occurrence;
+    TrainlogSessionExerciseInput assistance_occurrence;
+    TrainlogSessionExerciseInput continuous_occurrence;
+    TrainlogSessionExerciseInput large_occurrence;
+    TrainlogSessionExerciseInput temporal_occurrence;
+    TrainlogSessionExerciseInput temporal_max;
+    TrainlogSetInput first_sets[2] = {{8,0,false,0.0},{6,0,true,0.0}};
+    TrainlogSetInput second_sets[1] = {{11,0,true,30.0}};
+    TrainlogSetInput assistance_set[1] = {{9,0,true,15.0}};
+    TrainlogSetInput large_sets[65];
+    TrainlogTrainingExerciseContext context;
+    TrainlogExerciseOccurrence page[1];
+    TrainlogExerciseOccurrenceCursor next;
+    TrainlogExerciseOccurrenceCursor cursor;
+    bool has_more = false;
+    size_t count = 0U;
+    size_t index;
+    TrainlogOccurrenceSet set_page[64];
+    int next_position = -1;
+    char database_path[] = "/tmp/trainlog-context-XXXXXX";
+    int database_fd = mkstemp(database_path);
+    sqlite3 *raw = NULL;
+
+    CHECK(database_fd >= 0);
+    CHECK(close(database_fd) == 0);
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_insert_exercise_profiled(database, exercise_id, "Leg press renamed",
+        "leg press renamed", TRAINLOG_TRACKING_REPS, TRAINLOG_RECORDING_SETS, 0U) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_replace_exercise_body_zones(database, exercise_id, "thighs", zones, 1U)
+        == TRAINLOG_STATUS_OK);
+    base_occurrence(&occurrences[0], "entry_a", exercise_id, "leg_press", TRAINLOG_LOAD_EXTERNAL);
+    occurrences[0].sets=first_sets; occurrences[0].set_count=2U;
+    base_occurrence(&occurrences[1], "entry_b", exercise_id, "plate_loaded_leg_press", TRAINLOG_LOAD_EXTERNAL);
+    occurrences[1].sets=second_sets; occurrences[1].set_count=1U;
+    CHECK(insert_session(database,"session_b","2026-09-08T12:00:00+02:00",TRAINLOG_SESSION_TRAINING,occurrences,2U));
+    base_occurrence(&assistance_occurrence,"entry_assist",exercise_id,"leg_press",TRAINLOG_LOAD_ASSISTANCE);
+    assistance_occurrence.sets=assistance_set; assistance_occurrence.set_count=1U;
+    CHECK(insert_session(database,"session_a","2026-09-08T11:00:00+00:00",TRAINLOG_SESSION_TRAINING,&assistance_occurrence,1U));
+    base_occurrence(&max_occurrence,"entry_max",exercise_id,"leg_press",TRAINLOG_LOAD_EXTERNAL);
+    max_occurrence.has_max_weight=true; max_occurrence.max_weight_kg=120.0;
+    max_occurrence.target_sets=0; max_occurrence.target_reps=0;
+    max_occurrence.target_has_weight=false; max_occurrence.target_weight_kg=0.0;
+    max_occurrence.rest_seconds=0;
+    max_occurrence.load_mode=TRAINLOG_LOAD_NONE;
+    CHECK(insert_session(database,"session_max","2026-09-09T10:00:00+02:00",TRAINLOG_SESSION_MAX_TEST,&max_occurrence,1U));
+
+    CHECK(trainlog_training_exercise_context_load(database,exercise_id,4U,1U,&context)==TRAINLOG_STATUS_OK);
+    CHECK(strcmp(context.exercise.name,"Leg press renamed")==0);
+    CHECK(context.knowledge!=NULL && context.knowledge->interpretation!=NULL);
+    CHECK(context.persisted_zone_count==2U);
+    CHECK(context.compatible_equipment_count==2U);
+    CHECK(context.latest_max.found && context.latest_max.max_weight_kg==120.0);
+    CHECK(context.occurrence_count==4U);
+    CHECK(strcmp(context.occurrences[0].occurrence.entry_id,"entry_max")==0);
+    CHECK(context.occurrences[0].set_count==0U); /* explicit MAX is not a fake set */
+    CHECK(context.occurrences[1].occurrence.load_mode==TRAINLOG_LOAD_ASSISTANCE);
+    CHECK(strcmp(context.occurrences[2].occurrence.entry_id,"entry_b")==0);
+    CHECK(context.occurrences[2].set_count==1U && !context.occurrences[2].sets_have_more);
+    CHECK(strcmp(context.occurrences[3].occurrence.entry_id,"entry_a")==0);
+    CHECK(context.occurrences[3].set_count==1U && context.occurrences[3].sets_have_more);
+    CHECK(context.sets[context.occurrences[3].set_offset].has_weight==false);
+    trainlog_training_exercise_context_release(&context);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_a",0,64U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_OK);
+    CHECK(count==1U && !has_more && set_page[0].has_weight && set_page[0].weight_kg==0.0);
+
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_OK);
+    CHECK(count==1U && has_more);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&next,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_OK);
+    CHECK(count==1U);
+    cursor=next; cursor.entry_id[0]='\0';
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)memset(cursor.started_at,'x',sizeof(cursor.started_at));
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)memset(cursor.session_id,'x',sizeof(cursor.session_id));
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)memset(cursor.entry_id,'x',sizeof(cursor.entry_id));
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"not-a-date");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"2026-13-01T00:00:00+00:00");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    cursor=next; (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"2026-02-30T00:00:00+00:00");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+    {
+        const char *invalid_timestamp[] = {
+            "0000-01-01T00:00:00Z", "2026-09-05 18:34:12Z", "20260905T183412Z",
+            "2026-W36-5T18:34:12Z", "2026-09-05T18:34:12,5Z",
+            "2026-09-05T18:34:12+0200", "2026-09-05T18:34:12+02",
+            "2026-09-05T18:34.5Z", "2026-09-05T18:34:60Z", "2026-09-05T18:34:12+24:00"
+        };
+        for(index=0U;index<sizeof(invalid_timestamp)/sizeof(invalid_timestamp[0]);++index){
+            (void)memset(&cursor,0,sizeof(cursor));
+            (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"%s",invalid_timestamp[index]);
+            (void)snprintf(cursor.session_id,sizeof(cursor.session_id),"session_b");
+            (void)snprintf(cursor.entry_id,sizeof(cursor.entry_id),"entry_b");
+            CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+                &has_more,&next)==TRAINLOG_STATUS_INVALID_ARGUMENT);
+        }
+    }
+    cursor=next; (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"2026-09-08T10:00:00+15:00");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_OK);
+    (void)memset(&cursor,0,sizeof(cursor));
+    (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"2026-09-08T10:00:00Z");
+    (void)snprintf(cursor.session_id,sizeof(cursor.session_id),"session_b");
+    (void)snprintf(cursor.entry_id,sizeof(cursor.entry_id),"entry_b");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_OK);
+
+    /* TEMPORAL_READER_V1: production paging must preserve exceptional source
+     * spellings and use the same exact comparator for every exclusive cursor. */
+    base_occurrence(&temporal_occurrence,"entry_frac_low",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_frac_low","2026-09-21T10:00:00.1234567890123456Z",
+        TRAINLOG_SESSION_TRAINING,&temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_frac_high",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_frac_high","2026-09-21T10:00:00.1234567890123457Z",
+        TRAINLOG_SESSION_TRAINING,&temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_lower",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_lower","2026-09-20t10:00:00z",TRAINLOG_SESSION_TRAINING,
+        &temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_omit",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_omit","2026-09-22T10:00+15:00",TRAINLOG_SESSION_TRAINING,
+        &temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_high_offset",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_high_offset","2026-09-22T10:00:00+23:59",TRAINLOG_SESSION_TRAINING,
+        &temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_tie_a",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_tie_a","2026-09-20T12:00:00+02:00",TRAINLOG_SESSION_TRAINING,
+        &temporal_occurrence,1U));
+    base_occurrence(&temporal_occurrence,"entry_tie_z",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_occurrence.target_has_weight=false; temporal_occurrence.target_weight_kg=0.0;
+    CHECK(insert_session(database,"session_tie_z","2026-09-20T10:00:00-00:00",TRAINLOG_SESSION_TRAINING,
+        &temporal_occurrence,1U));
+    {
+        const char *expected[] = {"entry_omit","entry_high_offset","entry_frac_high","entry_frac_low",
+            "entry_tie_z","entry_tie_a","entry_lower"};
+        TrainlogExerciseOccurrenceCursor temporal_cursor;
+        const TrainlogExerciseOccurrenceCursor *after_temporal = NULL;
+        for(index=0U;index<7U;++index){
+            CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,after_temporal,1U,
+                page,&count,&has_more,&temporal_cursor)==TRAINLOG_STATUS_OK);
+            CHECK(count==1U && strcmp(page[0].entry_id,expected[index])==0);
+            CHECK(strcmp(page[0].started_at,temporal_cursor.started_at)==0);
+            after_temporal=&temporal_cursor;
+        }
+    }
+    base_occurrence(&temporal_max,"entry_max_offset",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_max.target_sets=0; temporal_max.target_reps=0; temporal_max.target_has_weight=false;
+    temporal_max.rest_seconds=0; temporal_max.has_max_weight=true; temporal_max.max_weight_kg=151.0;
+    CHECK(insert_session(database,"session_max_offset","2026-09-23T10:00:00+15:00",
+        TRAINLOG_SESSION_MAX_TEST,&temporal_max,1U));
+    base_occurrence(&temporal_max,"entry_max_true",exercise_id,"",TRAINLOG_LOAD_NONE);
+    temporal_max.target_sets=0; temporal_max.target_reps=0; temporal_max.target_has_weight=false;
+    temporal_max.rest_seconds=0; temporal_max.has_max_weight=true; temporal_max.max_weight_kg=152.0;
+    CHECK(insert_session(database,"session_max_true","2026-09-22T20:00:00Z",
+        TRAINLOG_SESSION_MAX_TEST,&temporal_max,1U));
+    CHECK(trainlog_database_latest_explicit_max_context(database,exercise_id,&context.latest_max)==TRAINLOG_STATUS_OK);
+    CHECK(context.latest_max.found && strcmp(context.latest_max.entry_id,"entry_max_true")==0 &&
+        context.latest_max.max_weight_kg==152.0);
+    (void)memset(&cursor,0,sizeof(cursor));
+    (void)snprintf(cursor.started_at,sizeof(cursor.started_at),"2026-09-08T12:00:00+02:00");
+    (void)snprintf(cursor.session_id,sizeof(cursor.session_id),"session_b");
+    (void)snprintf(cursor.entry_id,sizeof(cursor.entry_id),"entry_b");
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,&cursor,1U,page,&count,
+        &has_more,&next)==TRAINLOG_STATUS_OK);
+
+    CHECK(trainlog_database_insert_exercise(database,"ex_unknown_runtime","Custom renamed","custom renamed",
+        TRAINLOG_TRACKING_REPS)==TRAINLOG_STATUS_OK);
+    CHECK(trainlog_training_exercise_context_load(database,"ex_unknown_runtime",1U,1U,&context)==TRAINLOG_STATUS_OK);
+    CHECK(context.knowledge==NULL && context.occurrence_count==0U);
+    trainlog_training_exercise_context_release(&context);
+
+    CHECK(trainlog_database_insert_exercise_profiled(database,"ex_b1e6ffc6-75b5-45ff-a3c0-e7433c58013d",
+        "Marche renommée","marche renommee",TRAINLOG_TRACKING_DURATION,TRAINLOG_RECORDING_CONTINUOUS,
+        TRAINLOG_EXERCISE_DATA_SPEED_KMH)==TRAINLOG_STATUS_OK);
+    (void)memset(&continuous_occurrence,0,sizeof(continuous_occurrence));
+    (void)snprintf(continuous_occurrence.entry_id,sizeof(continuous_occurrence.entry_id),"entry_walk");
+    (void)snprintf(continuous_occurrence.exercise_id,sizeof(continuous_occurrence.exercise_id),"%s",
+        "ex_b1e6ffc6-75b5-45ff-a3c0-e7433c58013d");
+    (void)snprintf(continuous_occurrence.equipment_id,sizeof(continuous_occurrence.equipment_id),"treadmill");
+    continuous_occurrence.recording_mode=TRAINLOG_RECORDING_CONTINUOUS;
+    continuous_occurrence.data_fields=TRAINLOG_EXERCISE_DATA_SPEED_KMH;
+    continuous_occurrence.load_mode=TRAINLOG_LOAD_NONE;
+    continuous_occurrence.continuous_duration_seconds=900;
+    continuous_occurrence.continuous_has_speed=true;
+    continuous_occurrence.continuous_speed_kmh=5.0;
+    CHECK(insert_session(database,"session_walk","2026-09-06T10:00:00+02:00",TRAINLOG_SESSION_TRAINING,&continuous_occurrence,1U));
+    CHECK(trainlog_training_exercise_context_load(database,continuous_occurrence.exercise_id,1U,1U,&context)==TRAINLOG_STATUS_OK);
+    CHECK(context.occurrence_count==1U && context.occurrences[0].occurrence.continuous_duration_seconds==900);
+    CHECK(context.occurrences[0].set_count==0U);
+    trainlog_training_exercise_context_release(&context);
+
+    for(index=0U;index<65U;++index){large_sets[index].reps=(int)(index+1U);large_sets[index].duration_seconds=0;
+        large_sets[index].has_weight=false;large_sets[index].weight_kg=0.0;}
+    base_occurrence(&large_occurrence,"entry_large","ex_unknown_runtime","",TRAINLOG_LOAD_NONE);
+    large_occurrence.target_sets=65; large_occurrence.sets=large_sets; large_occurrence.set_count=65U;
+    CHECK(insert_session(database,"session_large","2026-09-05T10:00:00+02:00",TRAINLOG_SESSION_TRAINING,&large_occurrence,1U));
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",-1,64U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_OK);
+    CHECK(count==64U && has_more && next_position==63);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",next_position,64U,set_page,&count,
+        &has_more,&next_position)==TRAINLOG_STATUS_OK);
+    CHECK(count==1U && !has_more && set_page[0].position==64U);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='2026-09-20 10:00:00Z' WHERE session_id='session_lower';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,1U,page,&count,&has_more,
+        &next)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='2020-01-01T00:00:00.123456789012345678901234567890Z' WHERE session_id='session_lower';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,1U,page,&count,&has_more,
+        &next)==TRAINLOG_STATUS_OK);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='9999-12-31T23:59:59.123456789012345678901234567890Z' WHERE session_id='session_lower';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,1U,page,&count,&has_more,
+        &next)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='2026-09-20t10:00:00z' WHERE session_id='session_lower';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='9999-12-31T23:59:59.123456789012345678901234567890Z' WHERE session_id='session_max_true';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_latest_explicit_max_context(database,exercise_id,&context.latest_max)
+        == TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE sessions SET started_at='2026-09-22T20:00:00Z' WHERE session_id='session_max_true';",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE continuous_activity SET duration_seconds=2147483648 WHERE "
+        "session_exercise_row_id=(SELECT id FROM session_exercises WHERE entry_id='entry_walk');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,continuous_occurrence.exercise_id,NULL,1U,
+        page,&count,&has_more,&next)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE performed_sets SET reps=2147483648 WHERE position=0 AND "
+        "session_exercise_row_id=(SELECT id FROM session_exercises WHERE entry_id='entry_large');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",-1,64U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE performed_sets SET reps=0,duration_seconds=NULL,weight_kg=NULL WHERE position=0 AND "
+        "session_exercise_row_id=(SELECT id FROM session_exercises WHERE entry_id='entry_large');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",-1,1U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_OK);
+    CHECK(count==1U && set_page[0].has_reps && set_page[0].reps==0 && !set_page[0].has_duration &&
+        !set_page[0].has_weight);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "UPDATE performed_sets SET reps=NULL,duration_seconds='bad' WHERE position=0 AND "
+        "session_exercise_row_id=(SELECT id FROM session_exercises WHERE entry_id='entry_large');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",-1,64U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "PRAGMA ignore_check_constraints=ON; UPDATE performed_sets SET duration_seconds=NULL,position='bad'||position WHERE "
+        "session_exercise_row_id=(SELECT id FROM session_exercises WHERE entry_id='entry_large');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_occurrence_sets_page(database,"entry_large",-1,64U,set_page,&count,&has_more,
+        &next_position)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(sqlite3_open(database_path, &raw) == SQLITE_OK);
+    CHECK(sqlite3_exec(raw, "PRAGMA ignore_check_constraints=ON; UPDATE session_exercises SET data_fields='bad' WHERE "
+        "exercise_row_id=(SELECT id FROM exercises WHERE exercise_id='ex_b432623f-bfe9-4daf-a653-60ec7fdffbde');",
+        NULL, NULL, NULL) == SQLITE_OK);
+    CHECK(sqlite3_close(raw) == SQLITE_OK); raw = NULL;
+    CHECK(trainlog_database_open(database_path, &database) == TRAINLOG_STATUS_OK);
+    CHECK(trainlog_database_list_exercise_occurrences_page(database,exercise_id,NULL,1U,page,&count,&has_more,
+        &next)==TRAINLOG_STATUS_DATABASE_ERROR);
+    trainlog_database_close(database);
+    CHECK(unlink(database_path) == 0);
+    return 0;
+}

tui/tests/test_training_knowledge.c

@@ -0,0 +1,107 @@
+#include "trainlog/training_knowledge.h"
+
+#include <stdio.h>
+#include <string.h>
+
+#define CHECK(test) do { if (!(test)) { fprintf(stderr, "CHECK failed: %s:%d: %s\n", \
+    __FILE__, __LINE__, #test); return 1; } } while (0)
+
+static int contains(const TrainlogExerciseKnowledge *const *rows, size_t count, const char *id)
+{
+    size_t index;
+    for (index = 0; index < count; ++index) {
+        if (strcmp(rows[index]->exercise_id, id) == 0) return 1;
+    }
+    return 0;
+}
+
+int main(void)
+{
+    const char *const rear_delt = "ex_4cd2433e-80b1-478a-b8df-73fc6ef80962";
+    const char *const chest_press = "ex_8552dd77-fcd7-4f06-a1cc-d956eb1009af";
+    const char *const leg_press = "ex_b432623f-bfe9-4daf-a653-60ec7fdffbde";
+    const char *const rotary = "ex_1a34814c-2e46-40fc-b1f4-6d60b8e5a3e0";
+    const TrainlogExerciseKnowledge *rows[32];
+    const TrainlogExerciseKnowledge *record;
+    const TrainlogKnowledgeInterpretation *conditional;
+    TrainlogKnowledgeQuery query = {0};
+    size_t count = 0U;
+    size_t index;
+    int rotary_capability_found = 0;
+    const TrainlogKnowledgeBodyZoneAudit *audit;
+
+    CHECK(trainlog_knowledge_reference_count() == 23U);
+    CHECK(trainlog_knowledge_muscle_count() == 53U);
+    CHECK(trainlog_knowledge_joint_action_count() == 35U);
+    CHECK(trainlog_knowledge_movement_pattern_count() == 26U);
+    CHECK(trainlog_exercise_knowledge_count() == 23U);
+    CHECK(trainlog_equipment_knowledge_count() == 41U);
+    CHECK(trainlog_knowledge_body_zone_audit_count() == trainlog_exercise_knowledge_count());
+    audit = trainlog_knowledge_body_zone_audit_lookup(leg_press);
+    CHECK(audit != NULL && strcmp(audit->status, "confirmed") == 0);
+    CHECK(audit->rationale[0] != '\0' && audit->source_refs[0] != '\0');
+    audit = trainlog_knowledge_body_zone_audit_lookup(chest_press);
+    CHECK(audit != NULL && strcmp(audit->status, "questionable") == 0);
+    audit = trainlog_knowledge_body_zone_audit_lookup("ex_2d488c08-194c-4051-a3c9-34471646c1d3");
+    CHECK(audit != NULL && strcmp(audit->status, "unresolved") == 0);
+    CHECK(trainlog_knowledge_body_zone_audit_at(trainlog_knowledge_body_zone_audit_count()) == NULL);
+    CHECK(trainlog_knowledge_body_zone_audit_lookup(NULL) == NULL);
+    CHECK(trainlog_knowledge_muscle_lookup("pectoralis_major") != NULL);
+    CHECK(trainlog_knowledge_joint_action_lookup("shoulder_horizontal_abduction") != NULL);
+    CHECK(trainlog_knowledge_movement_pattern_lookup("horizontal_pull") != NULL);
+    CHECK(trainlog_knowledge_reference_lookup("openstax_upper") != NULL);
+    CHECK(trainlog_equipment_knowledge_lookup("rear_delt_pec_fly") != NULL);
+    CHECK(trainlog_exercise_knowledge_lookup("ex_unknown") == NULL);
+
+    record = trainlog_exercise_knowledge_lookup(rear_delt);
+    CHECK(record != NULL && record->interpretation != NULL);
+    CHECK(record->conditional_interpretation == NULL);
+    CHECK(strcmp(record->interpretation->primary_zone_id, "shoulders") == 0);
+
+    record = trainlog_exercise_knowledge_lookup(chest_press);
+    CHECK(record != NULL && record->interpretation == NULL);
+    conditional = trainlog_exercise_knowledge_conditional(chest_press);
+    CHECK(conditional != NULL);
+    CHECK(strcmp(conditional->family_description, "machine_chest_press") == 0);
+
+    query.muscle_id = "quadriceps";
+    query.muscle_role = TRAINLOG_KNOWLEDGE_ROLE_PRIMARY;
+    query.available_equipment_id = "leg_press";
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 32U, &count) == TRAINLOG_STATUS_OK);
+    CHECK(count == 1U && strcmp(rows[0]->exercise_id, leg_press) == 0);
+
+    query = (TrainlogKnowledgeQuery){0};
+    query.muscle_role = TRAINLOG_KNOWLEDGE_ROLE_PRIMARY;
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 32U, &count) ==
+          TRAINLOG_STATUS_INVALID_ARGUMENT);
+
+    query = (TrainlogKnowledgeQuery){0};
+    query.scientific_zone_id = "upper_body";
+    query.include_zone_descendants = true;
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 32U, &count) == TRAINLOG_STATUS_OK);
+    CHECK(contains(rows, count, rear_delt));
+    CHECK(!contains(rows, count, chest_press));
+
+    query = (TrainlogKnowledgeQuery){0};
+    query.movement_pattern_id = "horizontal_pull";
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 0U, &count) == TRAINLOG_STATUS_INVALID_ARGUMENT);
+    CHECK(count > 0U);
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 32U, &count) == TRAINLOG_STATUS_OK);
+    CHECK(count >= 2U);
+    query.movement_pattern_id = "not_a_pattern";
+    CHECK(trainlog_exercise_knowledge_query(&query, rows, 32U, &count) == TRAINLOG_STATUS_NOT_FOUND);
+
+    for (index = 0; index < trainlog_equipment_capability_count(); ++index) {
+        const TrainlogEquipmentCapability *capability = trainlog_equipment_capability_at(index);
+        if (strcmp(capability->equipment_id, "rotary_torso") == 0 &&
+                strstr(capability->exercise_ids, rotary) != NULL) {
+            rotary_capability_found = 1;
+            CHECK(strcmp(capability->link_status,
+                         "catalog_compatible_not_observed_occurrence") == 0);
+        }
+    }
+    /* Catalog compatibility is static metadata; the API contains no occurrence
+     * row and therefore cannot fabricate performance history. */
+    CHECK(rotary_capability_found);
+    return 0;
+}

tui/tests/test_tui_workflows.c

@@ -16,6 +16,10 @@ struct TrainlogTerminal {
     size_t event_index;
     char output[32768];
     size_t output_used;
+    int rows;
+    int columns;
+    bool coordinate_overflow;
+    bool text_overflow;
 };
 
 struct TrainlogPanel { int unused; };
@@ -29,12 +33,14 @@ static void script(TrainlogTerminal *terminal, const int *events, size_t count)
     (void)memset(terminal, 0, sizeof(*terminal));
     (void)memcpy(terminal->events, events, count * sizeof(events[0]));
     terminal->event_count = count;
+    terminal->rows = 24;
+    terminal->columns = 80;
 }
 
 TrainlogTerminal *trainlog_terminal_create(void) { return NULL; }
 void trainlog_terminal_destroy(TrainlogTerminal *terminal) { (void)terminal; }
-int trainlog_terminal_rows(const TrainlogTerminal *terminal) { (void)terminal; return 24; }
-int trainlog_terminal_columns(const TrainlogTerminal *terminal) { (void)terminal; return 80; }
+int trainlog_terminal_rows(const TrainlogTerminal *terminal) { return terminal->rows; }
+int trainlog_terminal_columns(const TrainlogTerminal *terminal) { return terminal->columns; }
 void trainlog_terminal_erase(TrainlogTerminal *terminal) { (void)terminal; }
 void trainlog_terminal_render(TrainlogTerminal *terminal) { (void)terminal; }
 void trainlog_terminal_style_on(TrainlogTerminal *terminal, TrainlogTextStyle style) { (void)terminal; (void)style; }
@@ -44,14 +50,28 @@ void trainlog_terminal_printf(TrainlogTerminal *terminal, int row, int column,
 {
     va_list arguments;
     int written;
-    (void)row;
-    (void)column;
+    if (row < 0 || row >= terminal->rows || column < 0 || column >= terminal->columns)
+        terminal->coordinate_overflow = true;
     if (terminal->output_used >= sizeof(terminal->output)) return;
     va_start(arguments, format);
     written = vsnprintf(terminal->output + terminal->output_used,
         sizeof(terminal->output) - terminal->output_used, format, arguments);
     va_end(arguments);
     if (written > 0 && (size_t)written < sizeof(terminal->output) - terminal->output_used) {
+        const char *text = terminal->output + terminal->output_used;
+        const char *at = text;
+        int cells = 0;
+        while (*at != '\0') {
+            utf8proc_int32_t codepoint;
+            utf8proc_ssize_t bytes = utf8proc_iterate((const utf8proc_uint8_t *)at,
+                -1, &codepoint);
+            int codepoint_cells;
+            if (bytes <= 0) { bytes = 1; codepoint = (unsigned char)*at; }
+            codepoint_cells = utf8proc_charwidth(codepoint);
+            cells += codepoint_cells < 0 ? 1 : codepoint_cells;
+            at += bytes;
+        }
+        if (column + cells >= terminal->columns) terminal->text_overflow = true;
         terminal->output_used += (size_t)written;
         terminal->output[terminal->output_used++] = '\n';
         terminal->output[terminal->output_used] = '\0';
@@ -240,12 +260,39 @@ static bool test_empty_sets_cannot_finish(void)
     return true;
 }
 
+static bool test_knowledge_scrolls_long_lists_at_minimum_terminal(void)
+{
+    TrainlogExercise exercise;
+    TrainlogTerminal terminal;
+    const int events[] = {
+        TRAINLOG_KEY_PAGE_DOWN, TRAINLOG_KEY_PAGE_DOWN, TRAINLOG_KEY_PAGE_DOWN,
+        TRAINLOG_KEY_PAGE_DOWN, 'b'
+    };
+
+    (void)memset(&exercise, 0, sizeof(exercise));
+    (void)snprintf(exercise.exercise_id, sizeof(exercise.exercise_id), "%s",
+        "ex_a72fa713-4b0e-431d-95e2-42d95beb77b1");
+    (void)snprintf(exercise.name, sizeof(exercise.name), "%s", "Lat pull");
+    script(&terminal, events, sizeof(events) / sizeof(events[0]));
+    terminal.rows = 20;
+    terminal.columns = 72;
+    tui_terminal = &terminal;
+    screen_exercise_knowledge(&exercise);
+    CHECK(!terminal.coordinate_overflow);
+    CHECK(!terminal.text_overflow);
+    CHECK(strstr(terminal.output, "Supra-épineux") != NULL);
+    CHECK(strstr(terminal.output, "Sources :") != NULL);
+    tui_terminal = NULL;
+    return true;
+}
+
 int main(void)
 {
     if (!test_assistance_creation_labels() ||
         !test_duration_creation_starts_empty() ||
         !test_append_requires_actual_and_rolls_back() ||
-        !test_empty_sets_cannot_finish()) return 1;
+        !test_empty_sets_cannot_finish() ||
+        !test_knowledge_scrolls_long_lists_at_minimum_terminal()) return 1;
     (void)printf("PASS tui_workflows\n");
     return 0;
 }