COUNCIL REVIEW: history inventory and physical diagnostics
Reviewer: history_inventory; snapshot HEAD b736c9df9491c2e50bff8a5c2bed834cc935f871.
Window: commits reachable from git --all since 2026-10-28T00:00:00+01:00, captured 2026-11-02T21:44:42Z.

INVENTORY AND COVERAGE
history_inventory.json records 72 refs, 241 unique commits (including three merges), dates, subjects, parents and per-commit changed paths with before/after blob IDs.
Standard git log --name-only reports 411 directly touched Markdown paths. 407 distinct Markdown paths were added by nonmerge commits (409 addition events across branches). Including files imported by comparing merge commits against EACH parent produces 734 paths, 806 distinct Markdown blobs, 8,574,371 bytes. These counts answer different questions; prefer 241 commits / 411 directly touched Markdown paths in the user-facing review.
All 806 blobs were read as bytes and archived, with SHA256 and line counts, under markdown_blobs/. This is mechanical inventory, NOT a claim of semantic reading of all 806 versions.
I read the full content of 60 paths listed in physics_reviewed_paths.json: all 42 Markdown files under release_research_implementation_20261031, two release_research_summary documents, physics review 00-09 plus RESULTS_VERIFIED, release plan00-04. Five v1 reports duplicate the corresponding v2 text, and are covered by identical content. Markdown hashes matching these reads are marked full_content_read_by_history_inventory in the manifest. Other council lanes and planning_docs cover different topics; do not infer their reading status from this inventory.
Generated monthly Git dossiers in physics_review/git contain older history, mechanically preserved but not reread line by line. Earlier historical source versions also remain inventory-only unless explicitly marked.
direct_markdown_changes.json gives exact latest changed commit/blob for each of the 411 paths. It does not claim that every branch tip has the same contents.

FINAL COUNCIL COVERAGE UPDATE
All411 paths are now accounted for in review_coverage.json:406 semantic/source-verified and5 generated historical inventories structurally verified. There are zero unreviewed latest-path blobs. Archive duplicates were content-deduplicated; generated result tables were checked against structured sources rather than treating repeated table lines as independent scientific content. For the five generated monthly Git dossiers, all1516 commit objects and section counts were verified and the156 messages in the requested recent window were read; pre-Oct28 historical metadata was not reread as new research. The manifest records the responsible lane, exact reviewed ref/blob and whether it matches the latest changed blob. The39 local branches,28 remote refs,4tags and1stash ref are pinned in history_inventory.json.
Additional lanes finished50 planning/general paths(52 versions),28 transfer dossiers,45 EvTTC paths,20 preregistration plans,71 runtime paths and137 detection paths; overlaps are deduplicated in the final union. README.md and docs/progress.md have divergent main/detection versions: both were read. The more recently committed detection branch is not a cumulative replacement for main. Full older versions elsewhere remain inventoried, not indiscriminately claimed as semantically read.

PRIORITY ARTICLE CONTRIBUTION
Include the dimensional/cadence intervention as a controlled diagnostic on the frozen deployed chain, together with its limits and negative downstream controls. This is stronger than presenting speculative physical auxiliaries as achieved improvements. The formulas themselves are classical projective kinematics, not a new physical law. The contribution is locating an interval-dependent response in PAIR, preserving the computation graph and testing population-level consequences.

METHODS DETAIL VERIFIED IN CODE
operational/phase_timewarp/canonical_pair.py first recomputes all three time-derived PAIR133 fields at canonical d0=0.1 s and reruns the frozen PAIR. For the canonical raw phase phi_c it computes q_c=-expm1(-phi_c)/d0; scales q*=q_c*d0/dt; returns phi*=-log1p(-d0*q*). Rows with dt exactly float32(0.1) preserve original bytes, bypassing the conversion. All batches, weights, normalizers and timing boundaries remain fixed. The corrected branch checks 1-d0*q*>0 and finite values. It reconstructs PHASE17 and recomputes contexts and fusion. Do not describe this as dividing the emitted TTC or scaling the original phase alone.
The deployed canonical phase phi=-log(1-d0/T), endpoint log-height change r, and physical rate q are distinct quantities. With h proportional to1/Z and constant axial speed, q_b=expm1(r)/dt. The r formula is not the same operation as interpreting phi. q(t+dt)=q/(1-dt*q) assumes constant axial speed in the chosen camera-depth coordinate, with valid pre-contact domain; it cannot anticipate arbitrary braking.

VERIFIED RESULTS (machine-readable values/keys/hashes in council_physics_claims.json)
1. Cadence population: docs/release_research_implementation_20261031/campaign_results_v1/POPULATION.json, routes.F0.domains.
Road: 2531/2531 paired, macro-by-recording MAE0.810620018160979 ->0.8104896119362653 s; relative error30.600389999028852 ->30.591045605308967 percent.
Slider:76/77 paired; MAE1.17815589826358 ->1.1208089503802752 s (-4.8675 percent); relative error83.885893959795 ->82.44204743312503 percent. No new missing or sign changes. The original missing Slider output stays missing.
Road delta MAE-0.00013040622471352978 s, acquisition-bootstrap95 percent CI[-0.0005047002743243974,+0.00009314967648063326], 10000 resamples seed7. See routes.F0.recording_bootstrap in the same JSON (inspect exact nested field before coding).
All 2459 newly added queries are road. Slider is one previously screened acquisition. Road interval includes zero; Slider improvement is not independently replicated. Numeric gate passed, no official promotion.

2. Modality localization on screen149: followup_results_v2/cadence_localization/METRICS.json and RESULTS.md. Road original/event-only0.748852855, RGB/both0.748672302 s. Slider original1.178155983, event1.163044442, RGB1.141746554, both1.120808948 s. This localizes the road change to RGB and shows contributions from both modalities in Slider. Frozen prescribed four-arm diagnosis; do not select a new model retrospectively. Nominal-event exact identities134 queries.

3. Native numerical sensitivity: campaign_results_v1/NATIVE_SENSITIVITY.json. Same-platform original/candidate5805/5805 physical queries; no new missing/sign changes, event context exact,86 exactly equal predictions. Absolute change median0.0000143051147461 s, p950.0000767707824707, p990.0001210403442383, max0.006496429443359375 s. These are prediction DIFFERENCES, not errors against truth.
Historical original parity remains failed:5805 intermediate comparisons and504 TTC tolerance failures. native_parity_distribution_v1/RESULTS.json gives historical p50/p95/p99/max0.000185489654542/0.00109405517578/0.00279678344727/0.496791839599609 s. Largest drift is in local predicted magnitude>16s. Four tested TF32 policies did not recover the profile parity. Do not attribute all drift to hardware as a proven cause, or call local repeatability historical reproducibility. No Test12 labels read; no score conservation established.

4. Observability controls: continued_results_v1/OBSERVABILITY.json.149 required/148 pairs,25 acquisitions; four ridge arms x leave-one-acquisition-out=100 fits. Macro gain-prediction MSE constant0.457492901, support0.492043720, physical0.919267453, permuted0.756780224. Physical-support delta0.427223732, CI[0.0805506904,1.0104172321]. This MSE is NOT TTC MAE. The gate is rejected. q algebraically estimable63/149;26 wrong signs;8 fits with zero residual degrees of freedom. Algebraic estimability does not certify model correctness.
Historical robust readout:8 q/TTC available,4 wrong signs; fallback MAE0.779543461 versus R0 0.766024827. OLS q63, TTC53, wrong signs26; fallback MAE1.593770789. Coverage must remain in the table.

5. Photometric identifiability fixture: followup_results_v2/sensor_identifiability/METRICS.json.24 constructed cases with identical event streams;12 have distinct q. Identity covers coordinates/polarity/timestamps to1ps, not merely count. Log-affine textures permit q-gradient product ambiguity; known reference reduces maximum recovered-q error to0.0337739915 s^-1. This limited constructed example is not a general impossibility theorem and does not calibrate the real sensor.
Real proxy:72 queries from24 road acquisitions; Slider77 excluded from photometric fit due provisional alignment, retained in inventory. Sensor reference control uses common mask and leave-acquisition-out calibration. RGB observed MSE96.657290 vs physical145.648563; difference CI[-83.9315,-0.2561]; versus activity136.458609, CI[-88.7005,+27.1035] includes zero. Residual is signed event-rate error, not TTC improvement. No identified contrast threshold/gain/reference/refractory calibration in the inventory.
Photometric q -> TTC: followup_results_v2/photometric_q/METRICS.json, readouts.<arm>.with_R0_fallback.candidate.macro_mae. Radial13.241014767456472; translation12.050232373047296; affine9.168523738352876; spatial permutation45.62959431247753; R0 baseline0.7660248271310539 s.72 valid q per arm,148/149 paired after fallback. Affine beats permutation yet severely worsens TTC: do not accept a sensor auxiliary merely because reconstruction improves. No producer/JEPA training opened.

6. Clock control after canonical PAIR: followup_results_v2/clock_cadence/METRICS.json, cadences.canonical.<arm>. Baseline EVENT_CONTEXT macro MAE1.9418902619679768 s; EWMA1.952822072, observer1.981365164, persistence/transport3.104231759. Their differences versus the matched context do not establish benefit. The synthetic dynamics bank verifies constant-speed exact transport but demonstrates delayed braking response; do not claim real improvement from the equation's correctness. Keep EVENT_CONTEXT and multimodal R0 baselines distinct.

7. Surface measurement: continued_results_v1/SURFACES.json.133/149 measurements,16 fallbacks, R0 absence preserved; MAE3.103982972 vs0.766024827 s, delta CI[0.934231309,4.198244944].146 histories auditable, zero changes of crop within triplets or between snapshots;53 constant anisotropic resizes. This rules out one zoom hypothesis for this cohort, not all occlusion/parallax errors.

8. Full local latency (separate from 5-second paced historical runtime): raw_latency_results_v1/RAW_LATENCY.json and PRECOMPUTED_LATENCY.json.120 first-ten-per-sequence queries, RTX5070Ti Laptop, warm weights, provided boxes, potentially warm OS cache. RAW original p50/p95/p99532.024/963.353/1014.155ms; canonical534.662/1008.124/1208.804ms (+4.6474 percent p95). NPZ-prepared original329.344/1603.654/1737.175ms; canonical346.744/1658.066/1791.578ms. Separate measurements: cannot subtract percentiles to infer preprocessing, or claim RAW intrinsically faster.120/120 RAW tensors exact, timed outputs exactly match local replay. Excludes detection, capture and physical transport. No sustained50Hz result.

EDITORIAL/SCIENTIFIC WARNINGS
- R0 is V16-derived TRAIN40 RGB+events, not V18 and not event-only despite historical head ID.1,955 parameters is only its fusion head, not nine-component system total.
- Three new gate seeds on shared producers are conditional replicates, not independent end-to-end retrainings or acquisitions.
- Old physics-plan tables cite423/434 labels and11 missing GT; this is superseded by Nov1 recovered434/434 evidence covered by council_evttc. Do not copy old numbers.
- results/RESULTS.md swapped RGB and event labels; use results_v2 and later followups, retain old report as negative provenance.
- followup_results_v1 says S1 pending; v2 supersedes it with executed proxy and negative TTC readout.
- Existing paper design_analysis.tex citation M_X05 for age-aware aggregation122.44 vs123.69 should be M_SIMPLEX (verify source registration); this is a citation correction, not a numerical change.
- Official scorer JSON does not embed prediction archive hash; avoid claiming cryptographic end-to-end binding beyond available receipts.
- No O2/D4/S2-S3/V2-V3/conditional-JEPA training occurred after the physics gates failed; implemented auxiliaries and126 tests are not achieved learned models.
- Treat all hypothesis dossiers as plans; do not include their proposed experiments in results.

RECOMMENDED ARTICLE PLACEMENT
Main: explicit temporal/phase contracts; cadence intervention table with populations and uncertainty; adapted-vs-frozen lineage table; one concise negative-control table (observability, clock, photometric, surfaces).
Appendix: exact derivations/domains, native numerical drift and latency boundaries, all raw metrics/provenance, chronological supersession and frozen protocols.
Exclude as novelty claims: classical q transport alone, unconstrained physical-state labels, physical-looking confidence, speculative JEPA reopening, untrained future architecture, safety/real-time/SOTA generalizations.
