trainlog/docs/reviews/session_generator_v1_scientific_delta_review.md

SESSION_GENERATOR_V1 final scientific delta review

Date: 2026-09-09. Baseline: fd955315ccc9bb835a13eb94d206f1e01d893dfb. Decision: SESSION_GENERATOR_V1_SCIENTIFIC_REVIEW=PASS. Reviewer: configured anatomie / Astra-high specialist, bounded final policy review. This clears the scientific policy gate for implementation; it does not declare implementation, executable validation or SESSION_GENERATOR_V1=PASS.

Scientific question

Does the reconciled final policy support an editable single-session suggestion without deriving prescription from repetition-less MAX, overstating observed performance, or interpreting authored exposure/selection rules as physiology?

Yes. No substantive scientific ambiguity remains that requires new research before encoding this bounded policy. Individual suitability and heuristic effectiveness remain explicitly uncertain; passing the review does not resolve those uncertainties through an unsupported efficacy claim.

Read the anatomy skill, originating scientific review, final domain and policy, approved architecture record, programming foundations and relevant existing reference records. Compared the domain and policy against retained original bytes in /tmp/trainlog-session-generator-v1/original-proposal/. Parsed policy comparison found changes only in numeric_rule_status, eligibility and load; goals, ranges, exposure, selection, duration, guidance and citations are unchanged. Architecture/storage/exchange choices were accepted as inputs, not independently re-reviewed here.

Sources

Existing evidence is reused; no new scientific reference or frozen catalog change is needed. Source verification was limited to the already cited papers.

• ACSM 2009: historical professional guidance supplies direction for repetition/rest conventions, not proof of the exact defaults or permission to substitute Trainlog MAX for a measured RM.
• Singer 2024: the rest synthesis reports uncertain, modest hypertrophy differences; it does not establish a universal 120-second optimum.
• Grgic and colleagues: reviewed comparisons do not generally require failure for adaptation. This does not establish equivalent stimulus from arbitrary easy sets.
• Vieira and colleagues: failure conditions produce greater acute fatigue in reviewed comparisons; logged repetitions and BODY ZONES do not measure those experimental outcomes.
• Kassiano and colleagues: limited intervention evidence supports considering systematic variation and cautions against excessive rotation. No exact Trainlog score or recency interval was validated by this review.

The five PubMed abstracts above were readable during this delta review. The originating review's summaries of ACSM 2026, Currier 2023 and Moran-Navarro 2017 are reused within their recorded limits: adult multiweek adaptation evidence and a small acute recovery experiment do not validate this algorithm. Their direct PubMed opens returned no readable body during this review; no fresh full-text verification is claimed. No commercial or EMG evidence underlies this delta.

Anatomical finding

No muscle-role or anatomical identity change is proposed. Existing resolved exercise interpretations remain the eligibility authority. A BODY ZONE is a projection of anatomy, not an interchangeable muscle dose or exercise identity. No additional anatomy discovery or persisted mapping migration is required.

Biomechanical finding

The retained exact exercise, exact non-null equipment, compatible external resistance and no-known-execution-conflict conditions are appropriate minimum boundaries for reusing a displayed load. Matching IDs do not establish identical settings, range, technique or calibration. Assistance and bodyweight do not become interchangeable external kilograms. Preserving an observed number unchanged avoids inventing an available increment or calibrated resistance. This applies the existing MAX/context foundation.

Exercise interpretation

Goal defaults/rest — accepted. General 2x10/90s, strength 3x6/180s, hypertrophy 3x10/120s and local muscular endurance 2x16/60s, with their unchanged editable ranges, are defensible authored templates. They are not RM tests, exclusive adaptation bands, weekly prescriptions or optimal individual doses. A strength label without characterized relative load promises no strength-specific intensity. Rest can be extended, affecting actual duration.

Load anchor — accepted. Within the inclusive 28-day window, select the newest qualifying completed occurrence in the required context. At least the proposed set count must contain positive finite actual weight and repetitions at least equal to the proposed repetitions. Return the minimum weight across all qualifying rows unchanged. Do not pool inadequate occurrences into an invented qualifying session. An increased target dose must be requalified.

This is a historical performance anchor, not a demonstration of today's capacity. Qualifying subsets may coexist with nonqualifying rows; warmups may qualify. The minimum can produce an insufficiently challenging suggestion. Actual rest, tempo, exercise order and accumulated fatigue are not established by the qualifying fields. A newer nonqualifying performance does not erase an older eligible observation, but the older observation cannot be described as verified current performance or automatic progression. Source and uncertainty must remain visible. These are limitations of the accepted rule, not new qualification criteria or a request to infer missing measurements.

MAX and equipment — accepted. An explicit MAX without repetitions supplies neither 1RM nor goal-specific repetition capacity. Without a qualifying actual dose, numeric target weight is absent for every goal. MAX may provide context only, using explicit_max_present_no_numeric_prescription; it cannot break equipment ties or supply a percentage fallback. Context choice follows newest qualifying actual anchor, then bytewise equipment identity. This review supersedes the originating review's acceptance of 50% MAX and its obsolete numeric-MAX-fallback test requirements.

Legacy representation clarification — accepted. The parent relayed the advisor's final clarification: legacy Android/V2 actual-only occurrences store plan load_mode=none. That value alone establishes no resistance semantics. Such a row may supply an anchor only with all targets absent and rest zero, when the exact known scientifically compatible exercise/equipment context positively establishes external resistance and actual positive weights satisfy every other anchor condition. Explicit external occurrences remain eligible; assisted, unknown, missing or conflicting contexts remain excluded. This bridge does not infer external resistance from a positive number alone, rewrite history or weaken the scientific repeated-dose requirement. A generated plan with absent weight may likewise use plan mode none while preserving distinct equipment resistance semantics. Parent-owned architecture/policy wording will record this representation clarification; no additional research is required.

Scoring — accepted as practical inference. Same-identity and same-pattern penalties can favor an alternative even when the familiar exercise has a load anchor. Their interaction with coverage priorities and the history bonus is explicit and deterministic. Neither repeating within those windows nor choosing an unanchored alternative is physiologically prohibited. Repeated use of the generator is not a validated rotation schedule or longitudinal strength program; specificity and familiar practice can be useful. No score is an effect size, readiness measure or injury probability. The unchanged diversity cap and duration estimate remain software conventions with explicit shortages.

BODY ZONE mapping

No mapping changes. Preserve separate primary and secondary actual-row counts, with primary precedence and per-row deduplication for grouped requests. Thresholds 1/3 primary/secondary within 24 hours and 6/12 within 72 hours are informational exposure conventions. Primary threshold gives warning; secondary-only threshold gives notice; otherwise none only after successful analysis. These labels are not clinical severity or measured recovery. Missing or unclassified history remains insufficient information. MAX-only and continuous activity exclusion means no comparable counted resistance-set dose, not absence of fatigue from those activities. Invalid required history analysis must not masquerade as no observed exposure.

Confidence

• high: established anatomy/context distinctions and the conclusion that repetition-less MAX cannot establish 1RM or repetition capacity.
• moderate: broad goal-template interpretation as an editable suggestion.
• uncertain: today's load/effort/readiness, exact scoring effectiveness, optimal volume/rest, and outcomes from repeated generator use.

Limitations

This is a final policy delta review, not new broad anatomy research, a clinical prescription review, or an implementation audit. Retained literature concerns population interventions or controlled acute protocols. None validates the 28-day freshness cutoff, exposure thresholds, scoring coefficients or duration formula as personalized physiology. No frozen knowledge payload was changed.

Implementation handoff

Implement the reconciled policy and architecture; no scientific numeric or eligibility wording correction is required. Preserve the limitations above in interpretation and explanations. Executable evidence should cover qualifying and insufficient actual rows, target-dose edits, all-qualifying-row minimum, newer nonqualifying versus older qualifying occurrence, exact context exclusion, inclusive 28-day age, absent MAX-derived loads and no invented rounding. Retain planned-versus-performed separation and existing exact temporal, exposure/aggregation, deterministic selection and shortage checks.

Two documentation-only updates belong to the parent, outside this review's write scope: update the stale policy status value that still says architecture is pending, and link this delta decision from the current domain/status record. Preserve the originating review as proposal history, with its MAX conclusion clearly superseded; do not let its former numeric-MAX tests drive implementation.

Reviewed policy SHA-256 before the representation clarification above: 912ba800d885916279b5f26a4923f8a39350de69c0c7ff021bfde3ec740d6fe2. Reviewed domain SHA-256 before the representation clarification above: beffc9a4ed97cd887a2ccd983e22a906dfa507b98c680083d41cee4bee5ecf50. Only this new review artifact was written. No production, policy, database, device, staging, commit or push operation was performed.