# Source, timing and inspection limits

**Film:** Introducing ChatGPT Study Mode, official OpenAI upload `XDYilxy1dn8`,
published July 29, 2025 according to retained official metadata. This is not a
fresh watch-page metadata refresh.

Source link: `https://www.youtube.com/watch?v=XDYilxy1dn8`

Selected source frames: **[115,153)**, 38 frames. Nominal source rate 30/1;
nominal interval 3.833333…–5.100000 seconds. The retained source-master SHA-256 is
`42dab89d7579b9e742a018291063b5e573bc3f3dbf61423e10e95c06543a9785`.

## Three different clocks/assets

1. **Native source evidence:** 38 retained full-context 1920×1080 PNGs, not
   source assets for this implementation. Source frame 115 is local 0. The
   separate retained PTS ledger gives 3.833 seconds at frame 115, 5.067 at frame
   152, and 5.100 at exclusive boundary 153. Its millisecond decimal entries
   alternate 33/34 ms; the original stream time base was 1/16000. The decimal
   ledger is not a recovery of every original tick or a claim of exact CFR.
2. **Separate research clip:** the retained silent source excerpt is 960×540,
   38 frames at exact 30/1 CFR with track time base 1/30 and duration 38/30
   seconds. It is a reduced-resolution research playback derivative, not the
   native source clock. Source footage is excluded from the public kit/ZIP.
3. **Both Remotion outputs:** 1920×1080, 38 frames at exact 30/1 CFR. Local
   frame n has presentation time n/30. No frame is dropped, duplicated or
   retimed to match the millisecond PTS jitter. The last frame starts at 37/30;
   the exclusive end is 38/30 = 1.266666… seconds. Both r01/r02 candidate movies
   actually use mux time base 1/15360, with 512 ticks per frame, as probed; this
   differs from the 1/30 source derivative time base without changing the CFR clock.

The native PTS window is 1.267 seconds at the ledger's precision, 0.333… ms
longer than the CFR window. `source-clock.json` records all 38 rows and the
exclusive end separately. Frame-index alignment is exact; equality of all
native source presentation timestamps with a CFR derivative is not claimed.

## Static inspection and revision limits

The builder examined all source local frames 0–37 and both retained outputs in
complete-frame consecutive sheets, then checked native-size book crops and new
render/plate comparisons. This is **static inspection, not perceived moving
playback**. Frame enumeration, a successful render or browser playback does not
establish a moving aesthetic judgment or a blind win.

Historical review02 noted broader leaf/hinge and early-cover differences. The
later frozen review06, inspected before r02, preferred the source to default
and tied source/adaptation under **static-only** inspection. Its actionable
default defect was a short shoulder at the raised left leaf's lower join,
primarily locals26–30 and secondarily12/16/22/24, native ROI [890,550,100,100].
Adaptation did not show that shelf. Straight versus bowed alone was not a
defensible failure in that review.

This revision tests one default lower-corner tangent repair. It preserves all
other geometry and the adaptation, including the early-cover states. All38
new default states were inspected as native-size static crops against source
and r01; adaptation's actual new movie and all38 PNGs match r01 byte-for-byte.
The25→26 and30→31 boundaries have separate numerical checks. A fresh independent
critic must judge the new stable movies; the builder does not claim a win. The work adds
one usable guide candidate for an existing mechanism, **zero new mechanisms,
zero source-fidelity completions and zero native Resolve verification credits**.

The original full-length source master is absent from this checkout; this run
uses the retained all-38-frame native evidence, PTS ledger and short excerpt.
No full-film moving inspection, redownload or original-master re-probe is claimed.

## Tool documentation

Preparation follows the installed Remotion 4.0.520 API, including its CDP
response wrapper. Full-browser selection is documented at
`https://www.remotion.dev/docs/renderer/open-browser`.

The Resolve guide targets the Fusion SVG and Media Pool image-sequence workflows
documented in the Fusion 19 reference manual:
`https://documents.blackmagicdesign.com/UserManuals/Fusion19_Manual.pdf`.
Menu locations and import fidelity still need workstation verification. This
documentation reference is not evidence that Resolve ran during this build.