# Source, clock and native gaps

**Source ID:** `DupfnOCH-JI` — OpenAI, *Work smarter with your company knowledge
in ChatGPT*. Retained publication metadata records **October 23, 2025**,
inside the inclusive **September 5, 2024–September 5, 2026** window.
The date basis is YouTube's watch-page `playerMicroformatRenderer.publishDate`,
retrieved September 5, 2026 at 01:20:32 UTC. This is retained-metadata
verification, not a new claim about a live web page.

The selected recipe and native archive index independently bind the exact
zero-based source interval **[240,360)** at **60/1 fps**, 1920×1080:
**[4.000,6.000) seconds**. Local frame 0 means source frame 240. Frame 119
means source frame 359; frame 360 is excluded. No lead-in, tail hold, retiming,
audio, interpolation or invented source performance is added.

Retained source-master SHA-256:
`88ba469d8878090a68e97474a743107b7bc25a52c1adf0d520816ba2dc665077`.
The full master is **not available at the recipe's recorded workspace path**.
Its hash is a metadata binding, not a hash of a master reopened in this task.

## What is actually available

The selected native archive retains **92 source PNGs**, each verified against
its member SHA-256 and PNG IHDR as 1920×1080. It is not a complete native
sequence. Its 276 total members also include two older reconstruction frame
sets; those are not additional source evidence.

| Missing local positions | Missing absolute source positions |
| --- | --- |
| 3–9 | 243–249 |
| 15–22 | 255–262 |
| 42 | 282 |
| 81 | 321 |
| 93 | 333 |
| 99–103 | 339–343 |
| 109–113 | 349–353 |

The retained **silent 960×540 MP4 excerpt contains all 120 temporal positions**.
Its SHA-256 is
`5c8d1566e75530a1a67f80e44778f39f57ed39a8b76a68552518565b39ab68bd`.
FFprobe reports a 1/60 time base, 60/1 average and nominal rates, zero start,
2.000000-second duration and no audio. This lower-resolution re-encode cannot
restore the 28 missing full-resolution PNGs or the original master's PTS.
The rational source clock comes from retained source metadata, while excerpt
and candidate clocks are independently measured outputs.

## Builder observations and limits

Consecutive native source windows show a right-edge text withdrawal, a faint
initial at local38 and no label at39; a narrow shell with three changing rings;
growing empty space before the first icon; hard icon arrivals at71/73/75/78/85;
and continuing ring motion through119. Border, text and pointer are separate
layers. The pointer departs downward rather than following the shrinking text.

The builder inspected native contact sheets and six consecutive comparison
sheets covering all 120 positions of the excerpt and each actual candidate.
This is prior-informed builder inspection, **not the requested distinct fresh
critic** and not an attestation of human real-time perception of every frame.
The browser also played all three MP4s at 1× and recorded actual callback
clocks. Its source playback dropped one frame; see `VERIFICATION.md`.

**An observed mismatch remains visible:** the shorter adaptation label stays
complete later than the source label during early contraction. Chrome measures
the two candidate strings at about 707.4735 px and 625.8845 px, despite both
containing 17 characters. Identical mask tracks therefore do not imply
identical letter-loss cadence. The supplied inputs and core are intentionally
preserved; this kit does not claim to repair that source fit.

Source font, original spline construction, easing, shutter, border filter and
alpha falloff are not recovered. The inherited per-frame fits are candidate
numbers, not official source-application controls. New nonbranded artwork and
Geist are substitutions. Neither attractive output nor a mechanical pass
closes moving-source quality review.