You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(watch,play): audio snaps onto the clock past a 200 ms stretch - #8675
After a congestion stall, audio is back on the shared clock about 2.5 s after the outage starts, where it used to trail by up to 1.4 s for 15 to 26 s. This holds in the browser and natively.
Threshold: stretch now defaults to 200 ms (js/watchSTRETCH, <moq-watch stretch>, moq play --stretch). Zero still turns the stretch off. A buffered player still uses the larger of its lookahead and the threshold.
Snap onto the clock: audio trailing the clock past the threshold now skips onto the clock instead of back to the bound (ring-buffer.tsread, ring.rsskip). The test is now "past the threshold, or past one window when that is larger", where it used to be "threshold plus a window".
Snap both ways: audio leading the clock past the threshold now pads silence until the clock reaches it, with the stretch on too. Before, it padded only when stretch was 0 (ring-buffer.ts, ring.rspad). Without this, native AAC congestion snapped onto the clock at 27.0 s, the target then rose 243 → 2023 ms, and audio sat 1.78 s ahead for 8.5 s.
stretch sizes nothing else:
The audio subscription reaches the estimator's 2 s ceiling past the delay, whatever the stretch (latency.tsmaxDelay, moq play's REACH).
Natively, the decode cap and the hole fill (fill_max) follow that reach.
Native video waits its own delay on a missing group. video_max_delay is gone.
This also fixes --delay auto --stretch 0, which cut the budget to 50 ms and dropped the sink on any hole longer than that.
Harness: recover_ms now runs from a disturbance to audio back within one codec frame of where it rested in the 5 s before (that rest leaves out earlier disturbances), in either direction. Before, it compared against the window's p50 and wasn't tied to any disturbance. The capture no longer takes a stretch.
Spec and docs: doc/concept/audio-jitter.md's observation point and skip rules, doc/lib/js/watch.md, doc/bin/cli.md, js/watch/README.md, and the rustdoc all describe the new rule.
Quests: deletes the quest file and its links, and notes in lag rule's Plan that both rings already snap both ways.
Decisions
Taken without asking (headless):
Padding ahead is in scope. The epic's Plan sets the 200 ms threshold "both ways", but the quest's Goal names only the trailing skip. The measured native AAC regression above says the two halves have to land together, so this PR adds the pad instead of leaving it to the lag rule, which keeps it.
Skip test on the clock: lag > max(bound, window), where it was bound + window. The threshold is the threshold, and a zero bound still waits one window.
Unclocked native ring keeps its old rule: past its bound (three quarters of the ring's headroom) plus one window, it skips back to the bound. With no clock there is nothing to land on, and the window of slack keeps a level hovering at the bound from skipping a few frames every callback.
The 2 s reach is a local constant in each player (latency.tsREACH, play/args.rsREACH), each documented as the estimator's BUCKET * BUCKETS. Exporting it from the estimator would add public API in @moq/hang and moq-audio for one consumer each.
bus::MAX_STRETCH stays 2 s. It caps a positioned channel's bound and was never the players' default.
recover_ms: the rest is read over 5 s, the tolerance is one codec frame (Opus 20 ms, AAC 1024 samples), it counts both directions, and it leaves earlier disturbances out of the rest. A row whose resting point moves for good, like native Opus step, never "returns" and reads to the end of the window, the same on main.
Budgets:
Replay rows: re-measured exactly, with the reasons in the note.
Congestion trimmed_ms_per_min (and native Opus skips): raised by the note's rule (worst run × 1.25 + one hiccup) from two local runs, marked provisional until Nightly re-measures.
The note says a rise in trimmed audio "is a regression to fix rather than budget". Here the trim is the snap the epic decided, so I budgeted it and said so in the note.
Validation
just check passes on the branch, including moq-audio, @moq/watch, the audio-quality client, and the just rs play tests (277).
New and updated unit tests in both languages:
Without a clock, a level half a window past the bound plays on and two windows past skips back to the bound (without_a_clock_only_a_window_past_the_bound_skips). It failed with a 720-frame skip before the unclocked threshold was restored.
A skip lands on the clock past the threshold and not before it.
A playhead ahead past the bound pads until the clock arrives and drops nothing; within the bound it keeps playing.
The subscription's max delay stays at delay + 2 s as stretch moves (200 ms → 1.5 s → 0).
recovery() cases.
just test audio-quality, one 60 s run per row on a loaded x86_64 Linux developer host (debug build, seed 7), main (d6195e8) against this branch. Both sides used this branch's analyzer so recover_ms compares like with like.
Congestion rows, A/V lag p95 / time to return:
Chromium AAC: 1396.8 → 61.6 ms / 26.3 → 2.5 s
Chromium Opus: 1424.8 → 64.3 ms / 20.3 → 2.8 s
Native AAC: 1246.3 → 116.7 ms / 14.5 → 2.5 s
Native Opus: 826.9 → 117.5 ms / 24.0 → 3.8 s
The other rows move within single-run noise. The chromium A/V lag p50 of about 30 ms on every row, which fails its ceilings, is the same on main. It is the output latency that output latency owns.
Replays (--replays --enforce) pass against the re-measured budgets:
relay-mic-stall A/V lag p95 484.9 → 118.6 ms. It now skips the outage's trail and the target falls, and pads the rise.
rtt40 skips its 320 → 80 ms target fall once.
bursty-aac skips a 225 ms trail once.
Before and after rows
chromium (main → this branch)
row
skips/min
trimmed ms/min
underruns/min
stalled
A/V lag p50
A/V lag p95
recover ms
target p95
aac-44100-bursty
1.1 → 0
36.3 → 0
0
0
56.3 → 61.6
82.1 → 86.8
750.9 → 1501.4
343.2 → 323.2
aac-44100-congestion
1.1 → 3.3
114.3 → 2498.7
0
0.041 → 0.05
55.5 → 29.8
1396.8 → 61.6
26268.1 → 2505.7
643.2 → 663.2
aac-44100-fixed-250
0
0
0
0
28.5 → 27.7
48.5 → 48.2
501.6 → 500.4
250
aac-44100-mild
0
0
5.5 → 0
0
27.7 → 28.8
48.6 → 49.1
500.5 → 500.6
203.2 → 186.2
aac-44100-near-zero
0
0
1.1 → 0
0
29.5 → 28.1
52.4 → 48.5
1530.8 → 500.8
223.2 → 203.2
aac-44100-step
0
0
5.5 → 2.2
0.005 → 0
33.3 → 31.9
58.2 → 54.5
1001 → 750.6
223.2
aac-44100-wide
0
0
0 → 6.5
0 → 0.005
33.5 → 39.3
56.2 → 69.7
750.5 → 1751.3
283.2
opus-48000-bursty
0
0
0
0
36.1 → 35.4
50.9 → 53.3
750.8 → 1000.8
160 → 240
opus-48000-congestion
0 → 5.5
0 → 2697.2
0
0.041 → 0.05
46.8 → 34.5
1424.8 → 64.3
20262.4 → 2751.7
600 → 560
opus-48000-fixed-250
0
0
0
0
36.1 → 32.2
53.2 → 52.7
752.1 → 751.1
250
opus-48000-mild
0
0
0
0
32.7 → 33.6
54.7 → 55.9
501.3 → 750.4
80
opus-48000-near-zero
0
0
0
0
33.3 → 34.6
56.2 → 57.8
752.2 → 1001.4
60 → 120
opus-48000-step
0
0
10.9 → 7.6
0
40.2 → 42.8
67.7 → 69.7
1252.7 → 1000.5
100
opus-48000-wide
0
0
0
0
43.2 → 40.2
66.9 → 69.4
751 → 2503.4
280 → 360
native (main → this branch)
row
skips/min
trimmed ms/min
underruns/min
stalled
A/V lag p50
A/V lag p95
recover ms
target p95
aac-44100-bursty
0
0
0
0
50.1 → 46.1
58.8 → 53.1
4249.8 → 0
343.2 → 323.2
aac-44100-congestion
4.3 → 7.6
271.7 → 2870.4
912.4 → 961
0
-4.8 → -4.2
1246.3 → 116.7
14499.9 → 2500.2
2023.2 → 923.2
aac-44100-fixed-250
0
0
0
0
-6.9 → -9.5
-1.5 → -4.2
0
250
aac-44100-mild
0
0
0
0
-17.3 → -11.4
-11.3 → -5.4
0
186.2
aac-44100-near-zero
0
0
7.6 → 0
0
-8.7 → -11.8
-2.4 → -5.1
0
186.2
aac-44100-step
0
0
11.9 → 22.8
0
-7 → -13.4
32.9 → 14.8
13750.6 → 499.7
223.2
aac-44100-wide
0
0
4.3 → 0
0
25 → 14.6
35.1 → 20.6
0
283.2 → 263.2
opus-48000-bursty
0
0
16.3 → 0
0
-4.6 → -0.9
2 → 2.5
0
140 → 180
opus-48000-congestion
2.2 → 7.6
403.3 → 3532.3
995.8 → 1096.8
0
-0.6 → -3.2
826.9 → 117.5
24000.6 → 3750.1
1040 → 820
opus-48000-fixed-250
0
0
0
0
-7.3 → -6.4
-2.5 → -1.7
0
250
opus-48000-mild
0
0
0
0
-0.6 → 0
5.1 → 8.1
0 → 249.2
60 → 80
opus-48000-near-zero
6.5 → 0
108.1 → 0
32.4 → 0
0
-4.9 → -4.6
5.8 → -2.6
501 → 0
100 → 60
opus-48000-step
0
0
25 → 39.1
0
30.4 → 24.8
54.7 → 50.7
30505.5 → 30504.2
100
opus-48000-wide
0
0
0
0
28.3 → 19.8
39.3 → 25.8
249.2 → 1499.5
200 → 180
Public API
Behavior only, no !:
stretch defaults to 200 ms in @moq/watch (Sync, Player, <moq-watch>) and moq play --stretch.
Past the threshold, audio skips onto the clock and pads when ahead.
The audio subscription no longer follows stretch.
moq_audio::playback::Sink::due keeps its signature; only its documented skip and pad behavior changes.
Wire: none.
Caused by
moq-dev/moq#8507 - stopped trimming audio that trails the clock, so a stall's backlog was caught up at the accelerate's pace
moq-dev/moq#8532 - bounded that trail at a 2 s stretch default, which also sized the subscription, the native decode cap, and native video's wait
Follow-ups
None.
Models
Size: M
Implementer: Claude, claude-opus-5-5, unknown effort
Implementer: Claude Code, claude-opus-5-5, unknown effort (iterate, XS)
…lock
The stretch setting now sizes only that threshold: the audio subscription,
the native decode cap and hole fill reach the estimator's 2 s ceiling past
the delay, and native video waits its own delay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…the clock arrives
A target rise right after a snap left native AAC congestion 1.78 s ahead for
8.5 s; the threshold now holds both ways.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
kixelated
changed the title
Audio recovers from a congestion stall quickly
fix(watch,play): audio snaps onto the clock past a 200 ms stretch
Nov 13, 2026
File: No configuration file from this repository applied
Commit: Unavailable
Layers: Organization UI
Review profile: CHILL
Plan: Advanced
Run ID: 1e5640c1-2f1e-4803-8e07-8af643dbcbf7
Commits
Reviewing files that changed from the base of the PR and between 2155336 and dbfd700.
Files selected for processing: 1
rs/moq-audio/src/ring.rs
No actionable comments were generated in the recent review.
Walkthrough
The changes set the default audio stretch to 200 ms and separate subscription reach from stretch. Audio rings now skip trailing audio onto the clock and pad silence when audio leads the clock beyond the applicable threshold. The CLI and browser documentation and tests reflect these playback rules. The audio-quality analyzer now measures recovery against recent resting lag, and the related measurement budgets have changed.
Merge Risk: Minimal · previous review 21553
This change sets the default audio stretch to 200 ms and makes audio skip onto the clock or pad silence past that threshold. The subscription reach stays at 2 s past the delay. The supplied summaries show consistent code, tests, docs and budgets, and no remaining merge-blocking risk was identified.
Pre-merge checks: 4
Check name
Status
Explanation
Title check
The title "fix(watch,play): audio snaps onto the clock past a 200 ms stretch" matches the main change. The PR sets the default stretch to 200 ms in js/watch and moq play. Audio that trails the clock p…
Linked Issues check
Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check
Check skipped because no linked issues were found for this pull request.
Description check
The description is specific and on-topic. It explains the congestion catch-up quest, the 200 ms stretch default, the skip-onto-clock and pad-when-ahead rules, the decoupling of the audio subscription …
Full details: Title check
Explanation
The title "fix(watch,play): audio snaps onto the clock past a 200 ms stretch" matches the main change. The PR sets the default stretch to 200 ms in js/watch and moq play. Audio that trails the clock past the threshold now skips onto the clock, and audio that leads it pads silence. The title names the skip behavior and the stretch threshold, and it covers both packages the changes touch.
Full details: Description check
Explanation
The description is specific and on-topic. It explains the congestion catch-up quest, the 200 ms stretch default, the skip-onto-clock and pad-when-ahead rules, the decoupling of the audio subscription from stretch, the recover_ms harness change, the budget updates, and the validation runs. It matches the changeset summaries for the docs, watch, audio, CLI, and audio-quality harness files.
Finishing Touches
Simplify code
Commit to this branch
Create a new PR
Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.
File: No configuration file from this repository applied
Commit: Unavailable
Layers: Organization UI
Review profile: CHILL
Plan: Advanced
Run ID: f7f93546-dfd4-410b-8aa9-bb8a5c7c25b9
Commits
Reviewing files that changed from the base of the PR and between 9c4b4df and 2155336.
Files selected for processing: 33
doc/bin/cli.md
doc/concept/audio-jitter.md
doc/lib/js/watch.md
js/watch/README.md
js/watch/src/audio/decoder.test.ts
js/watch/src/audio/decoder.ts
js/watch/src/audio/latency.ts
js/watch/src/audio/playout.test.ts
js/watch/src/audio/replay.ts
js/watch/src/audio/ring-buffer.test.ts
js/watch/src/audio/ring-buffer.ts
js/watch/src/element.ts
js/watch/src/player.ts
js/watch/src/sync.ts
quest/m0/nightly/README.md
quest/m2/audio-playout/README.md
quest/m2/audio-playout/congestion-catch-up.md
quest/m2/audio-playout/lag-rule.md
rs/moq-audio/src/bus.rs
rs/moq-audio/src/playback/sink.rs
rs/moq-audio/src/ring.rs
rs/moq-cli/src/play/args.rs
rs/moq-cli/src/play/headless.rs
rs/moq-cli/src/play/media.rs
test/audio-quality/README.md
test/audio-quality/budgets.json
test/audio-quality/clients/js/src/analyze.test.ts
test/audio-quality/clients/js/src/analyze.ts
test/audio-quality/clients/js/src/capture.test.ts
test/audio-quality/clients/js/src/capture.ts
test/audio-quality/clients/js/src/probe.ts
test/audio-quality/clients/js/src/ptt.ts
test/audio-quality/clients/js/src/schema.ts
Files with no reviewable changes: 4
quest/m2/audio-playout/congestion-catch-up.md
test/audio-quality/clients/js/src/capture.ts
test/audio-quality/clients/js/src/probe.ts
test/audio-quality/clients/js/src/ptt.ts
Included review availability: This review used your included allowance. Your plan provides up to 4 included reviews per hour; 0 remain after this review.
The reason will be displayed to describe this comment to others. Learn more.
ℹ️ No critical issues — one rough edge inline.
Reviewed changes
Reviewed the full diff: the 200 ms snap both ways in both rings, the subscription reach decoupled from stretch, and the harness's new recover_ms.
200 ms default: STRETCH and moq play --stretch now default to 200 ms; zero still turns the stretch off.
Skip onto the clock: ring-buffer.tsread and ring.rsskip trigger past max(bound, window) and skip the whole lag, not back to the bound.
Pad when ahead: both rings now pad silence past max(bound, window) with the stretch on, holding until the clock arrives; Rust gates it on clocked instead of !stretching().
Reach decoupled: latency.tsmaxDelay and args.rsmax_delay use a local 2 s REACH, video_max_delay is gone, and native video subscribes at its own delay.
Harness: longestAway becomes recovery(), which measures against the median lag of the 5 s before each disturbance and seeds from the warmup; capture, probe, and ptt drop stretch.
Budgets and docs: replay rows re-measured, congestion trim ceilings raised (provisional), and docs, rustdoc, and quests updated to the new rule.
The reason will be displayed to describe this comment to others. Learn more.
ℹ️ Not approving yet: one earlier Pullfrog thread is still open and unaddressed. Nothing else came up.
Reviewed changes
Reviewed the full diff: the stretch threshold becomes a 200 ms snap in both rings, the audio subscription and native budgets no longer depend on it, and the audio-quality harness gains a disturbance-anchored recover_ms.
200 ms default: STRETCH, <moq-watch stretch>, and moq play --stretch now default to 200 ms, with the docs, READMEs, and rustdoc updated to match.
Skip onto the clock: on the clock, ring-buffer.tsread and ring.rsskip skip the whole lag once it passes max(bound, window).
Pad when ahead: both rings pad silence once the lead passes max(bound, window), whether the stretch is on or off, and pad until the lead reaches zero. JS reads this.#bound and Rust reads inputs.bound, and each is that ring's own stretch bound.
stretch sizes nothing else: latency.tsmaxDelay and moq play's max_delay/set_max_delay loop now reach delay + 2 s (REACH). bound() is now max(stretch, lookahead). video_max_delay is gone, so native video waits only video_delay().
Harness: recovery() replaces longestAway. It times each disturbance until audio is back within one codec frame of its median lag over the 5 s before, leaves earlier disturbances out of that median, and takes the starting median from the warmup.
Budgets and quests: the replay rows are re-measured and the congestion rows' trim ceilings are raised provisionally, with the note explaining both. The finished quest file and its links are removed.
Rust and JS match on the skip and pad thresholds, on the join (a zero bound before the first clocked read), and in their tests.
ℹ️ The open thread on the unclocked native skip is still unaddressed
The earlier thread on rs/moq-audio/src/ring.rs still holds at head 2155336. The unclocked branch now skips once lag > max(bound, window), which is just bound, where it used to wait for bound + window. Each skip is still lag - bound, so this ring can now skip a few frames on almost every callback. That contradicts Decision 3's claim that the unclocked ring keeps its old behavior. Either apply the suggestion in that thread or reply there with why the new threshold is intended. Then resolve the thread and this PR can be approved.
Restores main's unclocked threshold, which the snap onto the clock had
narrowed to the bound alone, so a level hovering there skipped a few
frames nearly every callback.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The reason will be displayed to describe this comment to others. Learn more.
✅ No new issues found. The open thread on the unclocked skip is addressed.
Reviewed changes
Reviewed dbfd700, the only commit since the prior review at 2155336.
Unclocked skip threshold restored: ring.rsskip now splits on clocked. On the clock it still skips the whole lag once it passes max(bound, window). Without a clock it waits for bound + window and skips back to the bound, as on main, so a level hovering at the bound no longer skips a few frames every callback.
Regression test: without_a_clock_only_a_window_past_the_bound_skips checks that half a window past the bound plays on, and that two windows past skips once back to within one read of the bound.
Rustdoc: skip's doc comment now describes the unclocked rule.
The reason will be displayed to describe this comment to others. Learn more.
✅ No new issues found.
Reviewed changes
I reviewed the whole branch diff: the 200 ms snap in both rings, the subscription and native budgets no longer sized by stretch, the reworked recover_ms analyzer, the re-measured budgets, and the docs and quest updates.
Default threshold: STRETCH and moq play --stretch now default to 200 ms. Zero still turns the stretch off, and a buffered player still uses the larger of the stretch and its lookahead (latency.tsbound).
Skip onto the clock: trailing audio now skips when lag > max(bound, window) and lands on the clock. Before, it skipped back to the bound. ring-buffer.tsread and ring.rsskip agree, and only the unclocked native ring keeps the old rule of skipping back to the bound once it is a window past it.
Pad ahead with the stretch on: pad now applies at the same threshold whenever the ring is clocked, where before it applied only with the stretch off. With stretch 0 the threshold is one window, the same as before.
stretch no longer sizes the subscription: maxDelay and native max_delay reach the delay plus a 2 s REACH. The native decode cap and fill_max follow that, and video_max_delay is removed, so native video now waits only its own delay.
recover_ms: recovery() reads where audio rested from the median lag over the 5 s before each disturbance, leaving earlier disturbances out. The tolerance is one codec frame and both directions count. The warmup sets the rest for a disturbance at the start of the window.
Budgets and docs: the replay rows are re-measured exactly, and the congestion trim ceilings are raised and marked provisional. The concept doc, rustdoc, README, and element docs describe the new rule.
…stall rule together
Both #8675 and this branch re-measured the relay-mic-stall replay row for
their own change alone; the row now holds what the two measure together.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
kixelated
changed the title
![pullfrog[bot]](https://avatars.githubusercontent.com/in/1768019?s=60&v=4){kind=small}
Finishes quest/m2/audio-playout/congestion-catch-up.md, the first step of the audio playout epic planned in moq-dev/moq#8663.
Summary
After a congestion stall, audio is back on the shared clock about 2.5 s after the outage starts, where it used to trail by up to 1.4 s for 15 to 26 s. This holds in the browser and natively.
stretchnow defaults to 200 ms (js/watchSTRETCH,<moq-watch stretch>,moq play --stretch). Zero still turns the stretch off. A buffered player still uses the larger of its lookahead and the threshold.ring-buffer.tsread,ring.rsskip). The test is now "past the threshold, or past one window when that is larger", where it used to be "threshold plus a window".stretchwas 0 (ring-buffer.ts,ring.rspad). Without this, native AAC congestion snapped onto the clock at 27.0 s, the target then rose 243 → 2023 ms, and audio sat 1.78 s ahead for 8.5 s.stretchsizes nothing else:latency.tsmaxDelay,moq play'sREACH).fill_max) follow that reach.video_max_delayis gone.--delay auto --stretch 0, which cut the budget to 50 ms and dropped the sink on any hole longer than that.recover_msnow runs from a disturbance to audio back within one codec frame of where it rested in the 5 s before (that rest leaves out earlier disturbances), in either direction. Before, it compared against the window's p50 and wasn't tied to any disturbance. The capture no longer takes astretch.doc/concept/audio-jitter.md's observation point and skip rules,doc/lib/js/watch.md,doc/bin/cli.md,js/watch/README.md, and the rustdoc all describe the new rule.Decisions
Taken without asking (headless):
lag > max(bound, window), where it wasbound + window. The threshold is the threshold, and a zero bound still waits one window.latency.tsREACH,play/args.rsREACH), each documented as the estimator'sBUCKET * BUCKETS. Exporting it from the estimator would add public API in@moq/hangandmoq-audiofor one consumer each.bus::MAX_STRETCHstays 2 s. It caps a positioned channel's bound and was never the players' default.recover_ms: the rest is read over 5 s, the tolerance is one codec frame (Opus 20 ms, AAC 1024 samples), it counts both directions, and it leaves earlier disturbances out of the rest. A row whose resting point moves for good, like native Opusstep, never "returns" and reads to the end of the window, the same onmain.trimmed_ms_per_min(and native Opus skips): raised by the note's rule (worst run × 1.25 + one hiccup) from two local runs, marked provisional until Nightly re-measures.Validation
just checkpasses on the branch, includingmoq-audio,@moq/watch, the audio-quality client, and thejust rs playtests (277).without_a_clock_only_a_window_past_the_bound_skips). It failed with a 720-frame skip before the unclocked threshold was restored.stretchmoves (200 ms → 1.5 s → 0).recovery()cases.just test audio-quality, one 60 s run per row on a loaded x86_64 Linux developer host (debug build, seed 7),main(d6195e8) against this branch. Both sides used this branch's analyzer sorecover_mscompares like with like.main. It is the output latency that output latency owns.--replays --enforce) pass against the re-measured budgets:relay-mic-stallA/V lag p95 484.9 → 118.6 ms. It now skips the outage's trail and the target falls, and pads the rise.rtt40skips its 320 → 80 ms target fall once.bursty-aacskips a 225 ms trail once.Before and after rows
chromium (main → this branch)
native (main → this branch)
Public API
Behavior only, no
!:stretchdefaults to 200 ms in@moq/watch(Sync,Player,<moq-watch>) andmoq play --stretch.stretch.moq_audio::playback::Sink::duekeeps its signature; only its documented skip and pad behavior changes.Wire: none.
Caused by
stretchdefault, which also sized the subscription, the native decode cap, and native video's waitFollow-ups
None.
Models
Size: M
Implementer: Claude, claude-opus-5-5, unknown effort
Implementer: Claude Code, claude-opus-5-5, unknown effort (iterate, XS)
Review
Rewritten by
quest reviewafter each round.Head: dbfd700
Base: origin/main at 6f2f3ff
Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1d82e-bb6a-7b61-8049-4270ea0e68b0, round 1Score: 9/10,
Code inspection found no actionable regressions in the reviewed branch diff; builds and tests were not run, as instructed.Rounds: 1; findings 0: fixed 0, refuted 0, open 0
Findings: none