# Coverage and reconciliation

Research update: September 5, 2026. This matrix records research coverage, not deployment status. The public corpus remains **100 films**; this work adds production contracts and original studies, not another film quota.

## Baseline reconciled before expansion

The public builder integrates motion, openings, logos, and lower thirds. The lower-third work, still being finalized at the start of this task, has seven cases and 342 consecutive source frames; its later acceptance record is retained as existing work, not claimed as a new discovery. Opening, logo and lower-third documents have separate publication acceptance records. A concurrent site-font update is also preserved; it does not change the licensed DejaVu fonts in the original study bundle. A deployed snapshot and a later local working bundle are not interchangeable; compare their publication manifests before releasing an update.

| Area | Already established | This playbook adds | Still uncertain / gate |
| --- | --- | --- | --- |
| Film and screen coverage | 100 official films; 1,075 editorial shot groups; 637 screen occurrences | A 19-record source ledger joining selected findings to 12 executable recipes | Purposive sample; shot groups are not frame-exact cut lists |
| Motion | 100 clips; 63 newer close studies; 12 bounded native checks; six original lab diagrams | Shared phase, layer, units, duration and state-handoff contracts; independent recipe renders | No recovered easing, spring parameters or authoring tools |
| Openings and typography | Six cases; 47 phase anchors; 233 consecutive frames; contrasting cut, reflow and staged departures | `title-interface`, original fixed-line mask; explicit title-read interval and manual cue | Real title/font choice and calibrated reaction time require the production review |
| Chapter changes | Existing title-reset examples and feature chapters | `chapter-reset`: exact same-task return, a state invariant and a runnable interruption/return | Phrase length and actual pause alignment come from the recording |
| Framing / camera | Native Codex intermediate motion versus Atlas adjacent-frame cut | `context-detail`: normalized pivot, crop budget, chosen ease, settled reading interval | Numerical ease is proposed, not fitted; capture-resolution limit remains real |
| Pointer / typing | Pointer-first navigation and different native/editorial typing treatments | `pointer-intent`, `type-submit`: dwell, fixed text baseline, commit point, cursor parking | Actual product action and latency must be captured; neutral demo is synthetic |
| Selection / local actions | Email and document selection studies | `selection-action`: highlight beneath glyphs; toolbar follows scope; no completion claim | Available menu actions require current runtime verification |
| Scroll / masks | Calendar and embedded-preview examples | `scroll-landmark`: stationary mask/header, named destination, no reading while moving | Native app scroll physics not recovered; final offset must be captured |
| Loading / result | Existing preview and diff studies | Consecutive preview inspection separates container arrival from a still-building chart; `progress-result` | Edited source wait is not product latency; exact moment a viewer can read is a judgment |
| Result holds | Existing finished-output examples and short proposed holds | `result-hold`: words-based floor, 640 × 360 inspection, pixel-stability assertions | Reading floor is a proposal, not a universal comprehension measurement |
| Object continuity | Thumbnail-to-composer and attachment-to-viewer evidence | `object-handoff`: identical asset/anchor across a hard cut, verified pixel-for-pixel | Other morphs need their own intermediate-frame evidence |
| Endings / logos | 13 comparisons; six deep cases; 390 consecutive frames | `quiet-ending`: original three-bar construction, quiet tail, shared-timeline inequalities | No source mark is licensed for reuse; actual wave/pause and end window remain unapproved |
| Lower thirds | Seven local deep cases; hard-on, delayed field and across-cut variants | `anchored-caption`: optional artifact identifier, fixed frame anchor and contrast backing | Not a default speaker overlay in a separate presentation; font file unknown |
| Narration / sync | Two-video workflow, single review, supplied recording as clock | Backward scheduling from spoken proof, conflict rules, worked non-footage timing example | No supplied recording or final setup is evaluated in these neutral studies |
| Reproducibility | Working source players and isolated motion lab experiments | Offline original-assets bundle, schema, negative tests, 12 renders and three composed sequences | Real application capture and actual Slides/OBS playback remain future production checks |

## Targeted source inspection

No new films were downloaded. Five windows from existing official-channel masters were decoded and hash-checked: document selection 82.5–83.75 s; preview 91.75–93.25 s; calendar scroll 46.75–48.25 s; chapter build 69.5–70.75 s; final Codex output 121–130 s. The first four produce 177 consecutive source frames in total. The last uses 18 nearest frames at 0.5-second targets. The inspection ledger records frame indices, source PTS and source checksums. Decoding an interval does not imply an exhaustive inspection of its full film.

The most useful new finding is in the preview window: loading remains at 92.416667 s, dashboard content becomes faintly visible by 92.583333 s, and a chart is still extending in frames through 93.083333 s. This is bounded visual evidence that **first content arrival and finished readable proof are different events**. It does not recover product load time or define an exact perceptual settlement threshold. The existing IDE-diff example contrasts a different kind of completed output without a growing chart.

Selection inspection shows a complete selection already present at 82.5 s and a local toolbar appearing over the following frames; it does not newly locate the selection's first onset. The calendar sequence distinguishes scrolling content from its fixed outer frame and inset. Chapter samples retain letter-by-letter construction, not a claim of a universal word cadence. The final Codex samples bracket the change from output to closing title between 127.503 and 128.003 s; this boundary is explicitly **single-example and sampled**, not frame-exact.

## What is not being claimed

No soundtrack or narration analysis of the source films; no official font identification; no exact studio easing, authoring software, compositor project or latency recovery; no exhaustive channel inventory; no use of scene-score candidates as an edit list; no verified current product feature from a historical film; no actual meeting synchronization from synthetic research sequences.

The public original bundle contains no internal application code, catalogs, customer recordings, private URLs or full reference films. Source excerpts and study datasets stay in their established credited research routes. The production examples contain only original fixtures and licensed fonts. Publishing this research does not authorize final presentation production.