docs: define calibration science v2 heterogeneous optics

docs/architecture/calibration_science_v2.md

@@ -0,0 +1,503 @@
+# Calibration Science v2 — heterogeneous optics and adaptive acquisition
+
+**Status: PLANNED — current additive design authority.**
+
+Calibration Science v1 remains `PASS/FROZEN`. Science v2 does not weaken,
+reinterpret or retroactively replace v1 evidence. It defines the next
+scientific generation required by the product requirement that a Lardon3D
+project may contain photographs from heterogeneous cameras, lenses, focal
+configurations and normal autofocus operation.
+
+No device-specific autofocus envelope is validated by this document. Numeric
+focus-domain limits require dedicated physical evidence.
+
+## 1. Product/science problem
+
+A photogrammetry product must not require the operator to remember a rigid
+exposure recipe or split one physical subject into artificial projects merely
+because acquisition state changed.
+
+The required user experience is:
+
+```text
+user frames and captures
+-> Lardon3D records/observes acquisition state
+-> Lardon3D measures actual image quality
+-> Lardon3D resolves exact calibration applicability
+-> Capture remains usable when quality is good
+-> reconstruction becomes READY only when calibration coverage is complete
+```
+
+A normal supported camera body or rectilinear lens is equipment data. A new
+brand/model name alone must not require source-code modification.
+
+Source changes are justified only by a genuinely unsupported transport,
+encoded representation, geometric camera model or required device-state
+mechanism.
+
+## 2. Compatibility with Science v1
+
+The following remain frozen compatibility contracts:
+
+```text
+CALIBRATION_SCIENCE_V1=PASS/FROZEN
+CALIBRATION_TOOLING_V1=PASS/FROZEN
+CALIBRATION_BOOTSTRAP_V1=PASS/FROZEN
+CALIBRATION_WORKFLOW_V1=PASS/FROZEN
+```
+
+Historical S21 and A6000 Engine Bay campaigns remain:
+
+```text
+CALIBRATION_UNAVAILABLE
+BLOCKED_BY_KNOWN_CALIBRATION_DATA
+```
+
+Science v2 does not authorize retro-calibration by inference.
+
+## 3. Fundamental separation of states
+
+Science v2 separates three concepts that must never be conflated.
+
+### 3.1 Capture existence
+
+A Capture is durable acquisition evidence. Unsupported or unresolved optics do
+not erase it.
+
+### 3.2 Photo quality
+
+Quality describes whether the decoded representation is useful image evidence.
+It is evaluated from the image itself and may include:
+
+- sharpness;
+- defocus;
+- motion blur;
+- black/white clipping;
+- usable shadow/highlight information;
+- contrast;
+- texture;
+- noise or local signal-to-noise evidence when implemented;
+- downstream feature usability.
+
+Photo Quality may recommend `GOOD`, `SUSPECT` or `REJECT` without assigning any
+calibration identity.
+
+### 3.3 Calibration readiness
+
+Calibration readiness answers whether a selected scientific representation has
+exact compatible geometric calibration evidence.
+
+Therefore this is valid:
+
+```text
+QUALITY=GOOD
+CALIBRATION=CALIBRATION_REQUIRED
+```
+
+The image remains in the project.
+
+## 4. Heterogeneous selected executions
+
+A single project and one selected scientific execution may contain images from:
+
+- different camera bodies;
+- different lenses;
+- different focal configurations of one zoom;
+- different validated focus states or focus applicability domains;
+- different validated aperture applicability states;
+- different acquisition sessions.
+
+The scientific requirement is per selected image:
+
+```text
+selected image
+-> exactly one compatible calibration
+```
+
+It is not:
+
+```text
+selected execution
+-> one global calibration
+```
+
+A truthful READY state requires complete calibration coverage of the exact
+selected-image set.
+
+The current Sparse calibration scope already represents members as
+`image_id -> calibration_id`; v2 should reuse that per-image model unless
+repository inspection proves a real missing invariant.
+
+## 5. No silent substitution
+
+Science v2 preserves:
+
+```text
+NO_SILENT_SCIENTIFIC_SUBSTITUTION
+```
+
+The following are forbidden unless a later explicit validated v2 transfer model
+authorizes them:
+
+- another camera body's calibration;
+- another lens's calibration;
+- a nearby focal calibration;
+- a guessed focus calibration;
+- a guessed aperture compatibility;
+- a guessed stabilization/crop/pipeline equivalence;
+- fuzzy metadata matching that silently changes scientific identity;
+- interpolation or extrapolation outside a validated model/domain.
+
+If exact compatibility cannot be proven:
+
+```text
+CALIBRATION_REQUIRED
+```
+
+If more than one valid choice remains unresolved:
+
+```text
+SELECTION_REQUIRED
+```
+
+## 6. Exposure and photometric controls
+
+Purely photometric controls are adaptive acquisition variables.
+
+Science v2 does not reject an image merely because of the numeric value of:
+
+- shutter speed;
+- ISO;
+- white balance;
+- exposure compensation;
+- other controls proven not to change image geometry.
+
+Examples such as `1/125`, `ISO 800` or `f/8` from historical campaigns are
+observations, not universal validity thresholds.
+
+For shutter/ISO/white balance, actual decoded-image evidence decides whether
+the result is usable. A faster shutter with higher ISO may be preferable to a
+blurred low-ISO image; a slower shutter may be valid when the camera is stable.
+
+Missing photometric metadata must not cause fabricated values.
+
+## 7. Aperture
+
+Aperture is not automatically classified as purely photometric because some
+lenses can exhibit focus shift or geometric changes with aperture.
+
+The product goal remains that the operator does not manually manage scientific
+grouping.
+
+Science v2 must allow either:
+
+1. one calibration applicability proven valid across an aperture domain; or
+2. distinct automatically classified aperture applicability states.
+
+Unknown compatibility produces `CALIBRATION_REQUIRED`, not Capture rejection.
+
+No aperture-domain threshold is frozen without physical validation.
+
+## 8. Focal length and zoom
+
+Multiple focal configurations may coexist in one project and one selected
+execution.
+
+Exact calibrated focal states are the baseline v2 mechanism.
+
+Electronic focal metadata may select an exact compatible state only when its
+identity and trust rules are satisfied.
+
+Manual lenses or missing metadata remain normal product paths through explicit
+bounded state assignment.
+
+Calibration at one focal state is never silently borrowed for another focal
+state.
+
+A future continuous focal transfer/interpolation model is allowed only if it is
+a separately versioned Science v2 model with fit-domain evidence, hold-out
+validation, deterministic evaluation and hard rejection outside its validated
+domain.
+
+## 9. Autofocus and focus applicability
+
+Autofocus is a normal acquisition mode in Science v2.
+
+Normal product use must not require the user to lock manual focus for every
+project merely to satisfy calibration.
+
+Science v2 supports the following scientific concepts:
+
+### 9.1 Exact focus state
+
+A calibration may apply to one exact observed/declared focus state.
+
+This remains useful for manual lenses, fixed-focus systems and strict
+measurements.
+
+### 9.2 Validated focus applicability domain
+
+A calibration may apply across a bounded autofocus/focus domain only when
+physical evidence proves that applicability.
+
+Required evidence includes:
+
+- exact camera body and objective identities;
+- exact focal/geometric pipeline state;
+- explicit focus-state observation representation and provenance;
+- repeated physical calibration samples across the intended focus domain;
+- independent hold-out focus states;
+- comparison in image-space geometric error, not only parameter percentage;
+- bounded deterministic applicability;
+- explicit rejection outside the validated domain.
+
+Historical EXIF/MakerNote values such as Sony `FocusPosition2` or derived
+`FocusDistance2` may inform experiment design. They are not by themselves
+physical calibration evidence.
+
+### 9.3 Discrete focus bands
+
+If one calibration is not valid across the complete autofocus range, v2 may
+use multiple validated discrete focus bands or exact focus states.
+
+Automatic classification is required where trusted observed state permits it.
+
+Ambiguous overlap must not silently choose a calibration.
+
+### 9.4 Continuous focus transfer models
+
+Continuous interpolation across focus is not assumed.
+
+If future physical evidence justifies it, such a model must be:
+
+- explicitly versioned;
+- deterministic;
+- bounded to a validated domain;
+- fitted from retained physical evidence;
+- tested on excluded hold-out focus states;
+- rejected outside the domain;
+- incorporated into scientific identity.
+
+No extrapolation is silent.
+
+## 10. Stabilization, crop, orientation and processing pipeline
+
+Only state capable of changing geometry belongs to calibration applicability.
+
+Examples requiring explicit treatment unless equivalence is proven:
+
+- electronic stabilization;
+- geometric crop;
+- perspective correction;
+- resize;
+- computational geometry correction;
+- in-camera lens correction that changes pixel coordinates;
+- orientation normalization.
+
+Purely cosmetic color processing does not create a new geometric state after
+coordinate equivalence has been proven.
+
+Unknown geometric state is incompatible until resolved.
+
+## 11. Observed optical state and provenance
+
+Science v2 requires a bounded, versioned observed state sufficient to decide
+calibration applicability.
+
+Conceptually it may contain:
+
+```text
+camera_body_identity
+lens_identity
+focal_state
+focus_state_or_domain_observation
+aperture_state_when_geometrically_relevant
+stabilization_and_geometric_pipeline_state
+decoded_oriented_width_height
+crop_resize_transform_identity
+representation/decode_pipeline_identity
+observation_provenance
+```
+
+The exact public ABI and persistence layout are implementation-contract work.
+
+Important rules:
+
+- unknown is explicit;
+- unknown is never replaced with a guessed default;
+- electronic metadata is evidence with defined trust, not automatic truth for
+  every scientific field;
+- manual assignment is explicit and durable where metadata cannot identify a
+  state;
+- exact retry is idempotent.
+
+## 12. Automatic calibration resolution
+
+For each selected image:
+
+```text
+load exact Capture and representation
+-> load/derive trusted observed optical state
+-> enumerate validated compatible calibration applicability
+-> exactly one: calibration resolved
+-> none: CALIBRATION_REQUIRED
+-> multiple unresolved: SELECTION_REQUIRED
+```
+
+A selection must remain reviewable and durable.
+
+Calibration readiness does not rewrite Photo Quality evidence.
+
+## 13. Artifact / Tooling / Bootstrap evolution
+
+Science v2 does not require a new artifact merely because a project is
+heterogeneous.
+
+Implementation must first inspect and reuse existing primitives.
+
+Current architecture already has:
+
+```text
+SparseCalibrationScope member = image_id -> calibration_id
+L3DCALB1 entry = per-image intrinsics
+```
+
+However Tooling/Workflow v1 binds one global optical-state/provenance bundle.
+If heterogeneous independent calibration groups cannot be represented without
+losing provenance, an additive `L3DCALB2` / Tooling v2 / Bootstrap v2 may be
+defined.
+
+Any v2 artifact must preserve per-group:
+
+- optical/applicability identity;
+- target identity;
+- solver executable/configuration identity;
+- initialization evidence;
+- validation evidence;
+- model/dimensions;
+- exact published intrinsics;
+- exact selected-image membership.
+
+V1 artifact semantics remain unchanged.
+
+## 14. READY invariant
+
+A selected execution is READY only if:
+
+- every selected image is covered exactly once;
+- every calibration is scientifically compatible with that image's observed
+  geometric state;
+- the final calibration scope covers the exact selected image set;
+- no unresolved or ambiguous selected image remains;
+- publication/attachment follows the existing atomicity/idempotence contract.
+
+A project may retain any number of other non-selected or
+`CALIBRATION_REQUIRED` Captures.
+
+## 15. Device independence
+
+Normal equipment onboarding is data-driven.
+
+Required product properties:
+
+```text
+NEW_CAMERA_REQUIRES_CODE_CHANGE=NO
+NEW_LENS_REQUIRES_CODE_CHANGE=NO
+MULTIPLE_LENSES_PER_CAMERA=SUPPORTED
+ZOOM_MULTIPLE_FOCALS=SUPPORTED
+MULTIPLE_OPTICAL_CONFIGURATIONS_PER_PROJECT=SUPPORTED
+```
+
+A genuinely different geometric model, unsupported file format or unsupported
+transport may require new code. A different manufacturer/model string does not.
+
+## 16. Acquisition control philosophy
+
+Science v2 defines what must be observed and proven; it does not require one
+specific control transport.
+
+Future adapters should follow:
+
+```text
+if hardware permits control:
+    Lardon3D may choose/control settings automatically
+else if state can be observed:
+    Lardon3D verifies/classifies automatically
+else:
+    request only the minimum explicit user action required
+```
+
+HDMI, USB remote control and device-specific adapters remain separate future
+implementation layers.
+
+## 17. A6000 + E PZ 16-50 current evidence
+
+Historical A6000 engine-bay images provide operational design evidence only.
+
+Observed facts include:
+
+- E PZ 16-50mm F3.5-5.6 OSS;
+- heavy real use of 16-20 mm focal lengths;
+- AF-S across the historical JPEG set;
+- many distinct focus positions even within the 16 mm / f/8 subset;
+- exposure values varied with scene conditions.
+
+These observations demonstrate that a locked-focus single-state workflow is
+not an acceptable general product UX for this use case.
+
+They do not validate any reusable calibration or autofocus envelope.
+
+Current device-specific state:
+
+```text
+A6000_E_PZ_16_50_AF_APPLICABILITY=BLOCKED_BY_PHYSICAL_VALIDATION
+```
+
+## 18. Required future physical study
+
+Before claiming a real validated autofocus domain, capture dedicated ChArUco
+evidence across representative focus states and repeats while preserving the
+other relevant geometric state.
+
+The study must determine whether the lens supports:
+
+- one validated focus envelope;
+- multiple discrete focus bands;
+- exact focus-only calibration; or
+- a future explicitly validated continuous transfer model.
+
+Numeric domains and thresholds must come from the Science v2 validation design
+and retained physical evidence. They are not inferred from historical scene
+photos.
+
+## 19. Implementation dependency
+
+Current order:
+
+```text
+Calibration Science v2 design
+-> heterogeneous optical-state persistence/applicability foundation
+-> v2 Tooling/Bootstrap/artifact only if required
+-> heterogeneous Workflow v2 READY proof
+-> physical AF/optical applicability study
+-> dedicated calibrated real campaign
+-> real Sparse SfM
+```
+
+Do not start Sparse SfM or Dense/MVS as part of this design tranche.
+
+## 20. STOP conditions
+
+Stop rather than guess if:
+
+- a required geometric state cannot be observed or explicitly assigned;
+- a proposed compatibility relation lacks physical validation;
+- an AF/focal/aperture transfer model needs an invented threshold;
+- a new schema would force inference into historical rows;
+- heterogeneous calibration would require silently changing a FROZEN v1
+  scientific meaning;
+- a new camera requires a genuinely different geometric model not covered by
+  the current pinhole/distortion science.
+
+Uncertainty remains visible as `CALIBRATION_REQUIRED` or
+`SELECTION_REQUIRED`.

docs/roadmap/roadmap.md

@@ -24,7 +24,7 @@ SOURCE_COMMENT_AUDIT                  PASS
 PRODUCT_DEFINITION                    PASS/FROZEN
 PROMPT_TREE                           CURRENT
 USER_FACING_UI_LANGUAGE_NORMALIZATION PASS
-CURRENT_NEXT                          DEDICATED_PHYSICAL_CALIBRATED_REAL_CAMPAIGN
+CURRENT_NEXT                          CALIBRATION_SCIENCE_V2_DESIGN
 ```
 
 The current Project DB head is additive:
@@ -508,7 +508,28 @@ BLOCKED_BY_KNOWN_CALIBRATION_DATA
 
 S21 Engine Bay is permanently non-retro-calibrable under Calibration Science v1.
 
-Calibration Science v1 defines the physical protocol for future calibration acquisitions.
+Calibration Science v1 defines the frozen physical protocol for its compatibility path.
+
+Before the next physical campaign, the current dependency is Calibration Science v2. Real A6000
+engine-bay acquisition evidence established that normal operation must support autofocus and
+heterogeneous optical states without forcing the operator to maintain one locked focus or one optical
+configuration for the whole project.
+
+Science v2 is additive and must preserve these product invariants:
+
+```text
+one project may mix camera bodies, lenses and focal configurations
+one selected execution may use different calibrations per image
+unknown calibration state retains the Capture and reports CALIBRATION_REQUIRED
+shutter/ISO/white-balance values are adaptive, not image-validity constants
+image quality is judged from actual decoded-image evidence
+autofocus is a normal supported mode when its applicability is physically validated
+silent calibration substitution remains forbidden
+```
+
+The current Sparse calibration scope already models membership as `image_id -> calibration_id`; v2
+should reuse that capability rather than require one global calibration. Device-specific autofocus
+envelopes remain blocked until physical evidence validates them.
 
 Calibration Tooling v1 consumes an already acquired Science v1 evidence bundle, validates the bounded
 contract and produces deterministic `L3DCALB1` v1.

prompt/02_CURRENT_FROZEN_STATE.md

@@ -6,7 +6,7 @@
 CURRENT_PROJECT_DB_SCHEMA=v25
 PRODUCTION_TASK_KINDS=16
 USER_FACING_UI_LANGUAGE_NORMALIZATION=PASS
-CURRENT_IMPLEMENTATION_CURSOR=2_DEDICATED_PHYSICAL_CALIBRATED_REAL_CAMPAIGN
+CURRENT_IMPLEMENTATION_CURSOR=CALIBRATION_SCIENCE_V2_DESIGN
 ```
 
 ## Authority
@@ -35,6 +35,9 @@ Calibration workflow input boundary         PASS/FROZEN
 Calibration workflow evidence materialization PASS/FROZEN
 Calibration workflow selected-execution binding PASS/FROZEN
 Calibration workflow Tooling/Bootstrap READY PASS/FROZEN
+Calibration Science v2 heterogeneous optics  PLANNED
+Adaptive capture settings semantics          PLANNED
+Autofocus v2 foundation                      PLANNED
 Calibration Tooling planarity alignment     PASS/FROZEN
 ```
 
@@ -72,6 +75,35 @@ Preserve at their documented boundaries:
 
 `FROZEN` protects meaning and boundary, not necessarily every file byte.
 
+## Current calibration-v2 design intent
+
+Calibration v1 remains a valid frozen compatibility path. It is not weakened or
+retroactively reinterpreted.
+
+The current product dependency has changed before the physical campaign because
+real A6000 engine-bay acquisition evidence showed that normal operation requires
+heterogeneous optics and autofocus. Calibration Science v2 must therefore define
+an additive path in which:
+
+```text
+one project may mix camera bodies
+one project may mix lenses
+one project may mix focal configurations
+one selected execution may use multiple calibrations
+each selected image must resolve to exactly one compatible calibration
+autofocus is a normal supported acquisition mode
+exposure settings are adaptive rather than fixed scientific constants
+```
+
+Capture/quality acceptance and calibration readiness are distinct. A sharp,
+usable image with unresolved optics is retained and reported as
+`CALIBRATION_REQUIRED`; it is not discarded merely because calibration evidence
+is not yet available.
+
+No v2 rule may silently substitute another body, lens, focal state, focus state
+or calibration.
+
+
 ## Retained real S21 checkpoint
 
 ```text

prompt/12_CALIBRATION.md

@@ -16,14 +16,17 @@ CALIBRATION_WORKFLOW=PASS/FROZEN
 CALIBRATION_WORKFLOW_INPUT_BOUNDARY_V1=PASS/FROZEN
 CALIBRATION_WORKFLOW_EVIDENCE_MATERIALIZATION_V1=PASS/FROZEN
 CALIBRATION_WORKFLOW_SELECTED_EXECUTION_BINDING_V1=PASS/FROZEN
-CURRENT_CALIBRATION_NEXT=WORKFLOW_TOOLING_BOOTSTRAP_READY
 CALIBRATION_WORKFLOW_TOOLING_BOOTSTRAP_READY_V1=PASS/FROZEN
-CURRENT_CALIBRATION_NEXT=DEDICATED_PHYSICAL_CALIBRATED_REAL_CAMPAIGN
+CALIBRATION_SCIENCE_V2=PLANNED
+CALIBRATION_V2_HETEROGENEOUS_OPTICS=PLANNED
+ADAPTIVE_CAPTURE_SETTINGS_CONTRACT=PLANNED
+AUTOFOCUS_V2_FOUNDATION=PLANNED
+CURRENT_CALIBRATION_NEXT=CALIBRATION_SCIENCE_V2_DESIGN
 ```
 
 ## Authority
 
-`docs/architecture/calibration_science_v1.md`, `docs/architecture/calibration_bootstrap.md` and `docs/architecture/calibration_solver_preflight_v1.md` are the current specialized calibration documents.
+`docs/architecture/calibration_science_v1.md`, `docs/architecture/calibration_science_v2.md`, `docs/architecture/calibration_bootstrap.md` and `docs/architecture/calibration_solver_preflight_v1.md` are the current specialized calibration documents. Science v1 remains FROZEN; the v2 document is the current additive design authority until its contract is validated and frozen.
 
 `docs/architecture/calibration_tooling.md` is the specialized Tooling authority.
 
@@ -46,7 +49,65 @@ BLOCKED_BY_KNOWN_CALIBRATION_DATA
 
 Do not retro-calibrate them by invention.
 
-## FROZEN flow
+## Calibration Science v2 product requirements
+
+Science v2 is additive. It does not weaken or rewrite Science v1.
+
+The product requirement is:
+
+```text
+ONE_PROJECT_MAY_MIX_CAMERAS=YES
+ONE_PROJECT_MAY_MIX_LENSES=YES
+ONE_PROJECT_MAY_MIX_FOCAL_CONFIGURATIONS=YES
+PER_IMAGE_COMPATIBLE_CALIBRATION=REQUIRED
+
+AUTOFOCUS_NORMAL_OPERATION=SUPPORTED_TARGET
+MANUAL_FOCUS_LOCK_REQUIRED_FOR_NORMAL_CAPTURE=NO
+
+SHUTTER_SPEED_FIXED=NO
+ISO_FIXED=NO
+WHITE_BALANCE_FIXED=NO
+EXPOSURE_SETTINGS_ADAPTIVE=YES
+
+PHOTO_REJECTION_BASED_SOLELY_ON_EXPOSURE_METADATA=FORBIDDEN
+UNKNOWN_OPTICS_REJECT_CAPTURE=NO
+UNKNOWN_OPTICS_STATE=CALIBRATION_REQUIRED
+
+SILENT_CALIBRATION_SUBSTITUTION=FORBIDDEN
+```
+
+The core separation is:
+
+```text
+CAPTURE / PHOTO QUALITY
+!=
+CALIBRATION READINESS
+```
+
+A Capture may be sharp, non-clipped and scientifically useful as image data
+while still being `CALIBRATION_REQUIRED`. It remains durable in the project.
+
+Purely photometric settings such as shutter speed, ISO, white balance and
+exposure compensation may vary according to conditions. Their numeric values
+alone are not scientific rejection criteria. Actual decoded-image evidence
+remains authoritative for quality: sharpness/defocus/motion blur, clipping,
+usable tonal information, contrast/texture, noise or local SNR where available,
+and downstream feature usability.
+
+Camera/lens/focal/focus/aperture/stabilization/crop/pipeline state is treated
+according to whether it can alter geometry. Lardon3D must observe and classify
+that state, automatically select an exactly compatible validated calibration
+when one exists, expose `CALIBRATION_REQUIRED` when none exists, and expose
+`SELECTION_REQUIRED` when compatibility remains ambiguous.
+
+Autofocus is a normal v2 acquisition mode. A real autofocus applicability
+envelope must be supported only by physical calibration/hold-out evidence.
+Historical EXIF or MakerNote focus observations may guide experiment design but
+must never fabricate such evidence.
+
+See `docs/architecture/calibration_science_v2.md`.
+
+## FROZEN v1 flow
 
 ```text
 dedicated physical calibration acquisition

prompt/31_IMPLEMENTATION_ORDER.md

@@ -6,7 +6,7 @@
 IMPLEMENTATION_ORDER=DEPENDENCY_DRIVEN
 IMPLEMENTATION_AUTHORIZATION=NO
 STEP_0_USER_FACING_LANGUAGE_NORMALIZATION=PASS
-CURRENT_NEXT=2_DEDICATED_PHYSICAL_CALIBRATED_REAL_CAMPAIGN
+CURRENT_NEXT=CALIBRATION_SCIENCE_V2_DESIGN
 ```
 
 ## Authority
@@ -20,23 +20,25 @@ Implementation remains unauthorized until the human explicitly authorizes a tran
 Default dependency order:
 
 0. user-facing repository/UI language normalization where appropriate — PASS;
-1. final usable calibration workflow — PASS/FROZEN; valid external evidence now composes through FROZEN Tooling and Bootstrap to truthful READY;
+1. final usable Calibration Science/Tooling/Bootstrap/Workflow v1 compatibility path — PASS/FROZEN;
-2. dedicated physical calibrated real campaign;
+2. Calibration Science v2 design for heterogeneous cameras/lenses/focals, adaptive capture settings and autofocus — CURRENT;
-3. real Sparse SfM proof;
+3. Calibration v2 heterogeneous-optics persistence / Tooling / Bootstrap / workflow foundation;
-4. durable Dense/OpenMVS orchestration;
+4. physical autofocus/optical applicability validation and dedicated calibrated real campaign;
-5. mesh / refinement / texturing / export;
+5. real Sparse SfM proof;
-6. viewer foundation;
+6. durable Dense/OpenMVS orchestration;
-7. offline Coverage Analysis scientific contract and implementation;
+7. mesh / refinement / texturing / export;
-8. multi-campaign registration / fusion;
+8. viewer foundation;
-9. generic live acquisition adapter foundation;
+9. offline Coverage Analysis scientific contract and implementation;
-10. A6000 HDMI integration;
+10. multi-campaign registration / fusion;
-11. S21 integration;
+11. generic live acquisition adapter foundation;
-12. live camera localization;
+12. A6000 HDMI integration;
-13. live coverage overlay;
+13. S21 integration;
-14. actionable Capture Guidance;
+14. live camera localization;
-15. video ingestion / deterministic keyframes;
+15. live coverage overlay;
-16. final integration, UX, restart and performance proof;
+16. actionable Capture Guidance;
-17. Product Definition v1 Definition-of-Done closure.
+17. video ingestion / deterministic keyframes;
+18. final integration, UX, restart and performance proof;
+19. Product Definition v1 Definition-of-Done closure.
 
 ## Adjustment rule
 
@@ -45,3 +47,10 @@ Change this order only when current repository inspection proves a real dependen
 Step 0 closure does not authorize mechanical rewriting of historical persisted labels, path names, deliberate UTF-8 fixtures or persistence-sensitive internal strings. Such text is changed only when its owning scope requires it.
 
 Do not jump to A6000 live work merely because it is visually interesting. Reconstruction/viewer/coverage foundations required for truthful guidance come first.
+
+The calibration-v2 insertion before the physical campaign is dependency-driven,
+not feature creep: real acquisition evidence established that the v1 locked-focus
+single-optical-state path would make normal engine-bay capture impractical and
+would contradict the already-FROZEN product requirement for multiple optical
+configurations per project. Science v2 must be defined before collecting the
+physical evidence intended to validate it.