# Whole-body provisional coaching

Catalog: `track-photo-1`. Applies to every subject photo, paired or unpaired, regardless of session/year. Source roles select the subject; foreground bystanders and reference riders do not inherit the subject's findings.

## Automatic behavior

The app builds all nine topic rows for every subject, even with no notes file. Reviewed focus and rider-hypothesis rows appear directly; all observations, missing evidence and reusable practice cues remain available under All topics and practice cues. This supplements the existing three authored priorities; it does not discard topics to fit that limit.

This automates topic coverage and presentation. It does not run an image-language model or detect physical faults from the skeleton. New source-specific findings require visual assessment. Agents importing or reviewing photos must perform that assessment as part of the request, per CLAUDE.md, without a separate reminder from James.

## Review every topic

| Topic | Inspect | Do not infer |
|---|---|---|
| Wrist/grip | Hand orientation and room for arm movement | Tension or a locked grip from straight wrist alignment |
| Fingers/controls | Visible fingers, lever relationship; flag an on-bike access check if hidden | Always-two-finger clutch/brake coverage or unobstructed travel from a still |
| Inside elbow | Fold and down/out position with shoulder and hand | Universal numerical target or bar loading |
| Outside arm | Reach across tank and room for torso rotation | Locking, stiffness or causal restriction from extension alone |
| Head/torso | Hip hinge, chest/shoulder/head displacement and cornering context | Exact gaze, hidden rotation, braking/acceleration phase |
| Hips | Visible pelvis relative to seat and tank | Hidden seat contact, fixed hang-off distance |
| Knees | Inside opening, outside proximity, relationship to hip and foot | Rotation or pressure from a knee included angle |
| Foot angle | Toe down/out, visible clearance, control access context | Pavement contact from apparent proximity |
| Peg contact | Visible sole/peg relationship, possible ball-of-foot placement | Contact patch, pressure or constant foot placement when hidden |

Use `focus` for a supported useful cue, `observed` when no change is indicated, `hypothesis` for an attributed rider report or practice question, `unassessable` for missing evidence, and `notApplicable` with a reason when the topic does not apply. `needsReview` means unfinished; it is never a positive finding. A completed sweep has nine dispositions per subject, including visibility limits.

Inside/outside depends on the actual turn, not screen side or a CCW session name. The right hand still controls throttle and front brake in a right turn. No reference photo is required; use comparisons only when the corresponding reference and viewing context support them. Inspect originals before detector overlays and preserve watermarks.

## Authoring and audit

1. Inventory session manifests and identify source duplicates. Inspect each subject source; assess smaller copies on their own visibility limits.
2. For a saved review, `pose-batch coaching-checklist <review.json> <new-output.json>` creates a complete, source-bound pending template. Existing output is never overwritten.
3. Fill each topic with a disposition and evidence-based explanation. Record source-view turn direction and provenance. Save an array in `coaching-checklist.json` beside the manifest; back up existing content before replacing it. Preserve all source bindings. Keep these private files in ignored `data/`.
4. `pose-batch validate-checklist <review.json> <checklist.json>` checks schema/source validity and reports pending topics. `pose-batch audit-coaching <session.json> <results.json>` audits every subject, exits 2 for pending coverage and 1 for invalid input. Both use the app's shared catalog/validator.
5. Reopen the session and verify the correct subject's findings in Single and Compare. App loading rejects duplicate subjects/topics, missing topics, mismatched source/panel, reference-role entries, empty reviewed evidence and obsolete catalog versions. Invalid files reset findings to pending with a visible warning; other photo review remains usable. Conflicting turn overrides withhold the saved findings.

The validator verifies evidence binding and completeness, not coaching truth. Practice cues are provisional. General grounding: [ChampSchool body position](https://ridelikeachampion.com/why-body-position-matters-at-the-track/), [CanyonChasers control coverage](https://www.canyonchasers.net/2022/01/one-simple-habit-to-reduce-reaction-times/), [Honda riding position](https://global.honda/en/motorcycle-aftersales/plusone/202312.html). None validates an individual photo assessment.