[codeonym]

Summary

flow-reel record demo.ts now preserves each step’s original narration and writes SRT/WebVTT captions beside the silent MP4, using the final calibrated video timeline.

 record demo.ts
   demo.mp4
-  demo.transcript.json: step timings
+  demo.transcript.json: original narration + step timings
   demo.events.json
+  demo.srt
+  demo.vtt

One cue covers each step’s recorded boundaries. Captions normalize whitespace, let players wrap text, escape markup, and skip blank or zero-duration cues. Schema version 1 still accepts older transcripts without narration.

Evidence

• Before: the new recording test failed because every step’s narration was undefined; subtitle files were absent.
  After: both recording runs preserve narration, and ffprobe reads SRT cue boundaries matching each transcript.
• npm run typecheck and npm run build: passed.
• npm test: 45 passed.
• npm run test:e2e: 6 passed, including Chromium’s native WebVTT parsing, literal-text and active-cue playback checks, plus existing resolution, ±150 ms repeatability and two-frame alignment checks.
• Visually inspected the playback screenshot with literal markup, symbols and Unicode captions. Standards and specification reviews found no issues.

Merge Danger

Door: two-way

Blast Radius: recorder

Adds caption sidecars and narration fields; recording timing and silent MP4 behavior remain unchanged. Some SRT players display escape sequences rather than decoding them; the README documents this limit. WebVTT preserves literal text.

Closes #3