# Silent-viewer caption pacing for TrueTraceShorts

Use this when a proof short is understandable with voiceover but too fast or fragmented for viewers watching without sound.

## Trigger

The user reports that the text is too fast, sentence segmentation is awkward, or a single orphan word appears on the next subtitle page. Treat this as a blocking quality issue, not a minor styling preference.

## Principle

Captions must work as readable text, not only as active-word decoration. If a viewer watches on mute, they should still understand:

1. what the scam is,
2. what the red flag is,
3. what to do instead,
4. the memorable rule.

If this requires a longer video, extend the video. Do not squeeze comprehension into an arbitrary short runtime.

## Recommended repair pattern

1. Rewrite the script into short, complete sentences.
2. Render TTS sentence-by-sentence where possible.
3. Insert small inter-sentence pauses, e.g. ~650ms, to give captions breathing room.
4. Build caption pages from source-script sentence boundaries first, then map to forced-alignment word timestamps.
5. Prefer complete sentence or clause blocks over rigid word-count windows.
6. Avoid one-word and two-word orphan pages. If a sentence would leave a tiny tail, rebalance the block or allow a slightly longer caption page.
7. Merge aligner-split hyphenated words for display, e.g. `POP -UP` → `POP-UP`.
8. QA real rendered frames, especially mid-sentence and final-third frames, not only the contact sheet.

## Practical targets

- For simple/standard red flags, 35–45s is still the default, but silent-readable repairs may move into 45–60s.
- Reduce caption page count if the previous version felt frantic. Example: a 35.8s version with 26 caption pages was improved by extending to ~46s and reducing to ~15 caption pages.
- Keep active-word highlighting, but the static caption text must remain meaningful as a block.

## QA checklist

Before delivery, verify:

- no single orphan word on a new subtitle page;
- no sentence splice where a block crosses into the next idea awkwardly;
- captions are readable on mute;
- no clipping in the mobile safe zone;
- no black bars or squeezed 9:16 export;
- the caption text itself explains the red flag and safe action.

## Pitfall

Forced alignment can be technically correct while the viewer experience is poor. Timing accuracy is not enough; text pacing and semantic grouping must be designed for human reading.