GymCam rep grading

Everything stays on your machine — nothing is uploaded anywhere. When you finish (or pause), use Download CSV and send that file back.

Grading context

1. Your labels CSV (rep_labels_<you>.csv) — required 2. Clips folder (rep_clips/ or your clips folder) — required 3. Features file (features_<you>.json) — optional Enables the after-commit reveal of the pipeline's computed angles. Grading works fine without it. 4. Set-labels CSV (set_labels_<you>.csv) — optional If loaded, your per-clip (set-level) answers merge into it; otherwise a fresh one is generated. 5. AI predictions reveal (<context>__reveal.json, from grading/ai_reveal.py) — optional Enables "Verify with AI": a button proposes every mode's label up front for you to review, changed, and confirm. A hidden slice of reps get no button and grade exactly as blind, so tomorrow's labels can still feed fitting and validation.

standing reference

Verify with AI

Why the AI was wrong (optional)

confidence: High h Mod m Low l

Pipeline computed (revealed after commit — never before)

Blind AI prediction (revealed after commit — never before)

Disagreement is data — don't amend just because the AI says so. But if you re-examine the exact detail and conclude your call was wrong, amend: your blind first answer is kept and the change is logged as ai-amend.

Set-level: — did quality degrade across the set?

PA BU

1–4 set value, advance mode · ↑↓ move · d mode details · c copy prev rep · a use AI answers (verify mode) · h/m/l confidence · Enter commit / next · r replay · ,/. frame step · s speed · space pause