I found and fixed one wording error in `output/diagnosis.md` and made a few smaller fixes. The original is kept unchanged in `output/original-submission/`.

**Change log:**
1. **Overstated control:** step 4 and the operator reply said "Set **Export mode** to **Burn into video**". The captions screenshot (`03-captions.png`) shows "Export mode: Sidecar only" and, as a separate row, "Available option: Burn into video", plus an **Apply** button. It doesn't show an Export mode control you can change. Both now say "select **Burn into video**, then click **Apply**". The diagnosis also notes that the screenshot doesn't show how the option is chosen, and lists that under "Still to check".
2. **Unseen result stated as fact:** step 6 now says the Captions setting "should read" Burn into video.
3. **Missing sources:** each item under "Visible" now names its screenshot.
4. **Vague audio caveat:** it now asks plainly which clip plays at 00:00:42, since the screenshots don't show it.
5. **Word limit:** I tightened the wording to stay under 500 words after these changes.

**Checks performed:**
- **Original kept:** I copied the first version to `output/original-submission/diagnosis.md`, and a comparison confirmed it matches the pre-revision file exactly.
- **Evidence:** I re-checked every finding against the three screenshots, `product-help.md` and `operator-message.txt`. There were no calculations; times are copied as shown. The 00:01:00 length matches "60-second" and 16:9 at 1920 × 1080 matches "landscape". The sources don't conflict.
- **Requirements:** all are present:
  - the blocker and the warning;
  - numbered steps using the exact labels;
  - what changes after the fix;
  - what's visible kept separate from what still needs checking;
  - the operator reply.
- **Word count:** the revised file has 497 whitespace-separated tokens (475 actual words), under the 500 limit.
- **Labels:** I confirmed with a script that the key labels, error code, timestamp and file path appear in the revised file.
- **Claims:** neither the diagnosis nor the original handoff claims any action in the app.

**Files (in `output/`):**
- `diagnosis.md` (revised)
- `original-submission/diagnosis.md`
- `CHANGELOG.md`
- `wc.py`, `preserve.py`, `trim.py` (helper scripts, not deliverables)

**Remaining limitations:**
- **File and relink:** the interview file's current location isn't visible, and whether the relink works is untested.
- **Captions:** how Burn into video is actually chosen, and how the burned-in captions look, are unknown.
- **Audio:** whether the 00:00:42 peak sounds distorted, and which clip plays there, can't be told from the screenshots.
- **Export:** whether it completes and plays correctly is unverified, and there may be issues in panels not shown.

**Estimated human time (an estimate, not measured):** about 10 minutes to review and send the reply. The operator's fixes and export would take another 10–20 minutes.