reel-checklisted

# reel-check The video equivalent of plate: the last look before a reel leaves the kitchen. A reel leaks differently from prose, so it needs its own pass. plate scrubs a block of text you can read and edit. A reel is a rendered MP4 you **cannot grep**, built from two surfaces that each leak: 1. **The composition source** - the text that gets burned into the frames. 2. **The recording footage** - whatever the screen capture happened to show. You scrub the source, re-render or re-record, then verify the actual frames. The MP4 is the source of truth for what viewers see; source files can diverge from what got drawn. ## What to scan ### 1. Composition source (on-screen text) The captions, titles, and labels viewers read come from source files, not the video. Scan each: - **Cue / timeline JSON** - `captions[].text`, `titleCard.text`, `outroCard.text`, any label fields. - **Composition file** - `index.html` / `*.tsx` text nodes, `data-*` titles, alt text. - **Narration** - `SCRIPT.md`, TTS input text, voiceover scripts (a leak spoken aloud is still a leak). - **Brand sheet** - `DESIGN.md` and any generated metadata that ships with the project. Leak types are the same as plate: internal hostnames and machine names, private IPs (RFC 1918), LAN URLs / ports / dashboards, absolute home paths (`/home/<user>`, `/Users/<name>`), non-public personal emails, and AI-authorship disclosure. See plate for the full list, the doc-range replacements (RFC 5737 `192.0.2.x`), and the writing