Done — diagnosis written and within the word limit.

## Handoff

**Output:** `output/diagnosis.md` (498 words, limit 500 — counted programmatically via `check.py`)

**Contents:** immediate blocker (E204, offline `interview_camA.mp4`), secondary non-blocking warning (audio peak at 00:00:42), the Sidecar-only vs. burned-in captions gap driving the operator's actual requirement, 6 numbered steps using exact visible labels (Media, Relink, Captions, Export mode, Burn into video, Apply, Export), what changes after the fix, an explicit visible-evidence vs. needs-checking split, and a paste-ready operator reply. All factual claims cite the PNG filenames, `inputs/product-help.md` section headings, or `inputs/operator-message.txt`.

**Checks actually performed:**
- Read both text inputs in full; interpreted all three attached PNGs (panel text, statuses, paths, control labels).
- Verified word count ≤ 500 with a Python script (first draft was 502; trimmed to 498).
- Cross-checked that export settings (1920×1080, 16:9, MP4/H.264) already meet the operator's landscape constraint, so no aspect-ratio change is advised.

**Limitations:** This is screenshot interpretation only — I did not click, relink, export, or verify any outcome. The interview file's current disk location, relink match, audibility of the peak, and export success remain unverified and are flagged as such in the document. Raw inputs untouched.

**Estimated human finish time:** ~10–15 minutes (relink, toggle burn-in, listen at 00:00:42, export and spot-check the 60-second MP4). Estimate only, not measured.