E1: Retained measurement summary

Original summary SHA-256: f1529b4597deec65684b68f44d2bf61a795237c86927563043b6f15ae83d8362

Camera arrival and texture-read timing. 9 September 2026, Mountain Time. Apple M2 / 24 GiB.

The 5.003-second trace recorded 300 camera frame arrivals and 300 texture uploads.
The uploads read 229 distinct browser-frame timestamps. They repeated the preceding timestamp 71 times.
The pixel readback probe was off. This identifies repeated frame selection before rendering; it does not prove a final pixel delivery rate.

| Trace measurement | Mean, ms | Minimum, ms | 95th percentile, ms | Maximum, ms |
|---|---:|---:|---:|---:|
| Camera arrival interval | 16.674 | 9.042 | 21.933 | 24.200 |
| Texture upload interval | 16.659 | 13.373 | 18.773 | 19.824 |
| Enqueue to frame-slot update | 0.023 | 0.006 | 0.038 | 0.347 |

All browser media timestamp steps were 16.670 ms. Their actual arrival intervals varied from 9.042 to 24.200 ms.
Browser media timestamps are not native sample presentation timestamps. The trace cannot locate that timing conversion.

The exact tested Chromium source enables its smoothing buffer for remote tracks. The local camera path replaces a current-frame slot as frames arrive.
The texture upload reads that slot on its own schedule. The trace found exactly one frame read within each upload span.
All 300 arrivals reached the slot. Their mean handoff delay was 0.023 ms, and their maximum delay was 0.347 ms.
This sample therefore points to irregular arrival cadence rather than a large compositor handoff delay.

The native source audit found no confirmed timestamp-domain mismatch. Both native clock helpers use the host monotonic clock.
The sink discards the sender timestamp and stamps submission time. The extension retains the sample but supplies a fresh relay host time.
Its consume loop drains queued frames immediately, then waits at least 4 ms when empty. This is a possible jitter source.
The trace does not prove that this policy caused the measured jitter or certify the running extension's exact source revision.

The next diagnostic must compare a verified paced producer with the browser sender. That comparison can narrow the upstream boundary before a product edit.
The five-second trace cannot certify sustained performance. Native counts cover a slightly wider bracket than the trace markers.
No trace data loss occurred. Product sources stayed fixed, cleanup passed, and normal Sync was restored.
No product files, commits, or pushes changed.

Source evidence: Chromium revision `79460ebecaa5625e57a5fb679a735659e73dc687`, captured with hashes in `chromium-source.json`.
Relevant lines: `web_media_player_ms_compositor.cc` 249–258, 425–447, 781–816, 986–1061; `web_media_player_ms.cc` 1135–1139.
Native audit: `cmio_camera_sink.mm` 66–72, 229–237, 286–289; `camera_extension/main.mm` 302–309, 534–550.