From b124d9345e8e8eb5daf360d484dc4d7bf7abbd79 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 15 Nov 2026 06:21:53 -0800 Subject: [PATCH 1/3] chore(quest): claim quest/m2/audio-playout/replay-largest.md Quest-Claim: e7d40cab5bebcd6c From fd9641dd40e2d62678522311571dbc7e2ab62c4e Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 15 Nov 2026 06:41:19 -0800 Subject: [PATCH 2/3] test(audio-quality): the replay tells its consumer the publisher's largest group Each trace carries the publisher's largest group as the subscription began (version 4), and the replay answers it from the subscriber's largest() as a wire subscriber does, so the consumer reports the join's skipped backlog. relay-mic-rtt40's target p95 falls from 320 to 120 ms and its row is re-budgeted. The recorder and the capture observer record the field; the checked-in traces derive it. Co-Authored-By: Claude Sonnet 5.5 --- js/watch/src/audio/replay.test.ts | 23 ++++++- js/watch/src/audio/replay.ts | 47 +++++++++++-- quest/m2/audio-playout/replay-largest.md | 69 ------------------- test/audio-quality/README.md | 20 +++++- test/audio-quality/budgets.json | 18 ++--- test/audio-quality/clients/js/record.ts | 5 +- test/audio-quality/clients/js/replay.ts | 3 +- .../clients/js/src/capture.test.ts | 45 ++++++++++++ test/audio-quality/clients/js/src/capture.ts | 30 +++++++- test/audio-quality/clients/js/src/probe.ts | 1 + test/audio-quality/clients/js/src/recorder.ts | 17 ++++- test/audio-quality/clients/js/src/schema.ts | 9 ++- test/audio-quality/traces/bursty-aac.json | 5 +- test/audio-quality/traces/local-aac.json | 5 +- test/audio-quality/traces/relay-bbb.json | 5 +- .../audio-quality/traces/relay-mic-rtt40.json | 5 +- .../audio-quality/traces/relay-mic-stall.json | 5 +- test/audio-quality/traces/relay-mic.json | 5 +- 18 files changed, 212 insertions(+), 105 deletions(-) delete mode 100644 quest/m2/audio-playout/replay-largest.md diff --git a/js/watch/src/audio/replay.test.ts b/js/watch/src/audio/replay.test.ts index a32336d1fab..0297cdb37da 100644 --- a/js/watch/src/audio/replay.test.ts +++ b/js/watch/src/audio/replay.test.ts @@ -1,7 +1,8 @@ import { afterEach, beforeEach, describe, expect, it, type Mock, spyOn } from "bun:test"; import { Duration } from "@moq/clock"; import type * as Catalog from "@moq/hang/catalog"; -import { type Arrival, type Options, QUANTUM, type Quantum, replay } from "./replay"; +import { Track } from "@moq/net"; +import { type Arrival, answering, type Options, QUANTUM, type Quantum, replay } from "./replay"; const RATE = 48000; const CONFIG = { codec: "opus", sampleRate: RATE, numberOfChannels: 1 } as Catalog.AudioConfig; @@ -431,3 +432,23 @@ describe("buffered replay against the shared clock", () => { expect(warn).not.toHaveBeenCalled(); }); }); + +describe("answering", () => { + it("names the publisher's largest group first, then each newer group, and delegates the rest", async () => { + const track = new Track.Producer("audio"); + const subscriber = answering(track.subscribe(), 5n); + try { + // A group at or below the answer is no news: the track's own largest is older than it. + track.createGroup(3n); + expect(await subscriber.largest()).toEqual({ group: 5n, frame: 0 }); + const next = subscriber.largest(); + track.createGroup(5n); + track.createGroup(9n); + expect((await next)?.group).toBe(9n); + expect(subscriber.name).toBe("audio"); + } finally { + subscriber.close(); + track.close(); + } + }); +}); diff --git a/js/watch/src/audio/replay.ts b/js/watch/src/audio/replay.ts index 7756e1018c4..faf92d34c63 100644 --- a/js/watch/src/audio/replay.ts +++ b/js/watch/src/audio/replay.ts @@ -65,6 +65,12 @@ export interface Options { lookahead?: number; /** The configured stretch, in ms, as the element takes it (default: the element's). */ stretch?: number; + /** + * The publisher's largest group as the subscription began, which the consumer learns as a wire + * subscriber learns it from the publisher's answer. Omitted, the replay's track names its own: the + * first group to land. + */ + largest?: number; /** The rendition as the catalog described it: its advertised jitter, delay, and codec. */ config: Catalog.AudioConfig; /** @@ -177,6 +183,33 @@ const micros = (ms: number) => Math.round(ms * 1000); */ const settle = () => new Promise((resolve) => setImmediate(resolve)); +/** + * `track` answering `largest` the way a wire subscriber does once the publisher has answered: first + * the publisher's largest group as the subscription began, then each newer group as it lands. + * Everything else is the local subscriber's. + */ +export function answering(track: Track.Subscriber, largest: bigint): Track.Subscriber { + let answered = false; + return new Proxy(track, { + get(target, prop) { + if (prop !== "largest") { + const value = Reflect.get(target, prop, target); + return typeof value === "function" ? value.bind(target) : value; + } + return async () => { + if (!answered) { + answered = true; + return { group: largest, frame: 0 }; + } + for (;;) { + const next = await target.largest(); + if (!next || next.group > largest) return next; + } + }; + }, + }); +} + /** * Play `trace` in simulated real time, yielding every render quantum from the first arrival to the * end of the observation. @@ -218,11 +251,15 @@ export async function* replay(trace: Arrival[], options: Options): AsyncGenerato const maxDelay = new Derived([sync.out.delay, sync.out.maxDelay], (delay, max) => audioMaxDelay({ delay, maxDelay: max }), ); - const consumer = new Container.Consumer(track.subscribe({ delay: { max: WIRE_MAX_DELAY } }), { - format: new Container.Legacy.Format("audio"), - delay: new Derived([sync.out.delay, maxDelay], (min, max) => ({ min, max })), - wait: sync.out.maxDelay, - }); + const subscriber = track.subscribe({ delay: { max: WIRE_MAX_DELAY } }); + const consumer = new Container.Consumer( + options.largest === undefined ? subscriber : answering(subscriber, BigInt(options.largest)), + { + format: new Container.Legacy.Format("audio"), + delay: new Derived([sync.out.delay, maxDelay], (min, max) => ({ min, max })), + wait: sync.out.maxDelay, + }, + ); // Match the decoder: a codec constant until a delivered packet states its own duration. const signals = new Effect(); diff --git a/quest/m2/audio-playout/replay-largest.md b/quest/m2/audio-playout/replay-largest.md deleted file mode 100644 index 1b45561ae13..00000000000 --- a/quest/m2/audio-playout/replay-largest.md +++ /dev/null @@ -1,69 +0,0 @@ -# [S] Replay the publisher's largest group at tune-in - -## Goal - -The audio-quality replay measures the join's backlog the way a live player -does. Each recorded trace carries the publisher's largest group as the -subscription began, and the replay's consumer learns that number as a wire -subscriber learns it from the publisher's answer. `relay-mic-rtt40`'s target -p95 then falls from 320 ms to 120 ms, and its budget row tracks the consumers' -backlog reports, so a regression there fails the row. - -Boundaries: the native decode path's replay is still owed separately (see the -epic's README). No API is added to `@moq/net` or the container consumer for the -replay's sake, and the checked-in recordings are not re-recorded. - -## Plan - -Since [moq-dev/moq#8761](https://github.com/moq-dev/moq/pull/8761), both -consumers report a group they skip unread as join backlog when its sequence is -at or below the publisher's largest group at subscribe time. The replay -(`js/watch/src/audio/replay.ts`) writes a trace into a local -`Track.Producer`, whose largest comes from the first group to land. For -`relay-mic-rtt40` that is the stale tune-in frame, group 7499, below the -backlog the consumer closes, so the replay reports nothing and still measures -320 ms. - -Decided 2026-11-15 (maintainer: plan it in m2, under this epic). The rest was -settled from the trace and a local measurement: - -1. **The trace carries it.** `Trace` (`test/audio-quality/clients/js/src/schema.ts`) - gains a required `largest: number`, the publisher's largest group as the - subscription began, and its `version` goes to 4. Every checked-in trace gets - one, so no trace means "the track's own" by leaving it out. -2. **New captures record it.** The capture observer - (`test/audio-quality/clients/js/src/capture.ts`) reads its subscription's - `largest()` the way `js/hang`'s consumer does: once the track's info or - first group shows the subscription established and `start()` resolved, so a - cached group racing the answer, or live groups landing after it, don't set - it. `record.ts` writes it into the trace. -3. **Checked-in traces derive it.** None of them recorded the subscribe - answer, and re-recording would need a person at the MacBook's microphone and - would move every other column of the baseline. Each gets the newest group - whose media timestamp plus the path's lowest lag (the smallest - `at - timestamp` over the trace) falls at or before the first arrival: the - live edge as the first frame arrived. Its `source` names the derivation. - For `relay-mic-rtt40` that is group 8994. The backlog the relay held - reaches about 9047, and the live frames after it settle at that lag from - about 1.3 s in. -4. **The replay answers it.** The replay hands its consumer a subscriber whose - `largest()` answers the trace's `largest` first, as a wire subscriber's does - once the publisher answered, then each newer group as it lands. It - delegates everything else to the local subscriber. Rejected: creating an - empty group head at `largest`, which delivers a group the recording never - saw and moves the start; a consumer option, which is API for a test. -5. **Budgets follow.** Replay budgets equal what they measure, so every row - that moves is re-budgeted in the same PR, and the note in `budgets.json` - loses its sentence about the stale tune-in frame. Measured locally on - #8761's head with `largest` 8994 injected, `relay-mic-rtt40` reads target - p95 120 (from 320). It also moves underruns/min 0.3 to 0.7, player skips/min - 0.3 to 0, trimmed ms/min 83 to 0, accelerated ms/min 88.3 to 97.7, A/V lag - p50 0.6 to 2.3 and p95 6.9 to 8.6, and recover ms 250.7 to 749.3: the - trade #8761's pick 3 accepted. Group 9047 measured the same row, so the - derived number is not on a knife's edge. -6. **Docs.** `test/audio-quality/README.md`'s Traces section documents the - field, how a capture records it and how the old traces derive it. - -## Related - -- [moq-dev/moq#8761](https://github.com/moq-dev/moq/pull/8761) - the consumers' backlog reports this replay measures diff --git a/test/audio-quality/README.md b/test/audio-quality/README.md index d158f06fc89..0274cf1cc34 100644 --- a/test/audio-quality/README.md +++ b/test/audio-quality/README.md @@ -125,7 +125,8 @@ Chromium GC timeline in `.trace.json` and sampled CPU profile in `.pro Tracing can perturb playback; compare it with a separate untraced control and never use traced rows to calibrate quality budgets. -To replay a capture, put its arrivals and final environment configuration into the version 3 +To replay a capture, put its arrivals, final environment configuration, and `largest` (the +observer's join, `environment.largest` in `.page.json`) into the version 4 `Trace` below, subtract the first arrival's `at` from every arrival and from the last sample's `at` for the duration, and supply the observed minimum RTT as a diagnostic. Replay measures the target from the arrivals and holds the configuration constant; compare the raw samples when it @@ -215,7 +216,7 @@ them, each graded as a `replay` row whose profile is the file's name: | `relay-mic-stall` | `relay-mic` with nothing arriving from 15 s to 16.5 s, then the held frames trickling in oldest first at twice real time until they meet live arrivals at 18 s: the stall a congested path causes, longer than the target, with its backlog arriving late | A trace is a [`Trace`](clients/js/src/schema.ts): the catalog's audio rendition, the smallest -PROBE round trip the session saw, how long it observed, and every frame's `[at, timestamp, group, duration]`, +PROBE round trip the session saw, the publisher's largest group at tune-in, how long it observed, and every frame's `[at, timestamp, group, duration]`, with `at` on the viewer's clock and `duration` in ms. Duration comes from the container's sample metadata, the Opus packet TOC, or the codec's fixed sample count. Capture refuses a frame whose duration cannot be determined; timestamp gaps never stand in for missing packet payloads. @@ -224,7 +225,8 @@ receives it, with the same transport, container, and codec-duration readers over on the delay, so the replay decides those again. `replay.ts` plays a trace through the player's own `Container.Consumer` and ring on a simulated -clock ([`js/watch/src/audio/replay.ts`](../../js/watch/src/audio/replay.ts)), at the "auto" delay a +clock ([`js/watch/src/audio/replay.ts`](../../js/watch/src/audio/replay.ts)), whose consumer is told +the trace's `largest` as a wire subscriber is told by the publisher, at the "auto" delay a real `Sync` resolves from the playout target that consumer measures, and reads each quantum through the tap's classifier. The consumer makes the player's group ordering, max delay skips, and discontinuity resets again at that delay, and the ring follows the delay as it moves; a group's stream is taken to finish with its last recorded frame. Every @@ -252,6 +254,18 @@ just test audio-quality-record --lanes relay-mic --mic-broadcast microphone, instead of publishing Chromium's fake capture device. Name the browser and device in the trace's `source` before checking it in. +Version 4 adds `largest`, the publisher's largest group as the subscription began, which a wire +subscriber learns from the publisher's answer and the replay's consumer needs to tell a group it +skips unread as the join's backlog. The recorder and the `--capture` observer read it the way +`js/hang`'s consumer does: once the track's info or first group shows the subscription established +and its start resolved, so a cached group racing the answer, or live groups landing after it, do not +set it. No earlier recording kept the answer, and re-recording needs a person at the microphone and +would move every other column of the baseline, so each checked-in trace derives it: the newest group +whose first frame's media timestamp plus the trace's lowest lag (the smallest `at - timestamp`) falls +at or before the first arrival, the live edge as the first frame arrived. Its `source` says so. For +`relay-mic-rtt40` that is group 8994, where its first arrival is the stale group 7499 and the relay's +backlog reaches about 9047. + Version 3 adds per-frame durations. The AAC-LC recordings made before that field existed retain their arrival clocks and use the codec's fixed 1024 samples per frame. The Opus `relay-mic` recording is a real microphone captured with `--mic-broadcast`, with its packet durations retained. The diff --git a/test/audio-quality/budgets.json b/test/audio-quality/budgets.json index 75fad02e381..87bf5117bb4 100644 --- a/test/audio-quality/budgets.json +++ b/test/audio-quality/budgets.json @@ -1,5 +1,5 @@ { - "note": "The browser rows are measured on the nightly runner (ubuntu-24.04-arm, headless Chromium, seed 7, 60 s a row), from the audio-quality-run artifacts of the seven scheduled Nightly runs on main from 2026-10-29 to 2026-11-04 (40885347517, 41065371094, 41204380716, 41313466339, 41483654426, 41656613559, 41852517544), the first after a fresh subscriber stopped receiving the relay's cached history. The six AAC rows other than congestion are measured the same way from the seven runs from 2026-11-05 to 2026-11-11 (42050683248, 42236107954, 42377867523, 42487375545, 42659882487, 42842827718, 43041909115), the first after the TS importer published each PES as one group; the first four predate the browser's time stretch, and the six rows keep the four provisional ceilings set below from 2026-11-09 and 2026-11-10. Each ceiling is the worst of its seven nights times 1.25, plus one runner hiccup, rounded up to two significant figures. The hiccup is sized from the single gaps the 2026-10-08 night put on otherwise clean rows, the fixed-250 control among them: 250 underruns/min, 700 gap ms/min, 3.3 skips/min, 280 dropped ms/min (both sized on the playhead's skip-ahead inference, and now applied to player_skips_per_min and trimmed_ms_per_min), 0.01 stalled, 0.015 silence, and one 20 ms bucket of target. Budgeted this way, every night passes when it is held out of its row's measurement except one, and that row's own ceiling includes it: the 2026-11-03 Opus bursty row raised its target to 300 ms 2.6 s in and held it for one 20 s half-life before settling at 180. Such startup holds, Opus near-zero's 100 ms on 2026-10-30 and 10-31 among them, are budgeted as measured. These ceilings record today's player as the runner hears it, not target audio quality. To re-measure, download the same artifact from the latest scheduled nightlies (`gh run download -n audio-quality-run`), which keep every run, and apply the same rule. On 2026-11-10 the inferred skip_aheads and discarded_ms keys were dropped, since they read a time stretch's catch-up as dropped audio, and every browser row gained player_skips_per_min, trimmed_ms_per_min, and av_lag_ms p50 and p95, by the same rule from the only two scheduled nights on main with today's browser player and its trim counter, 2026-11-09 and 2026-11-10 (42659882487, 42842827718). A lag is signed, so its ceiling is the worst night plus a quarter of its size plus one 20 ms bucket. These four keys are provisional until seven nights with the stretch model re-measure every row. The grade follows the model: audio rests on the shared clock, trails it through a stall, and catches up after. When a change moves a row, a steady-state lag (av_lag_ms p50) that rises, or audio the player trimmed, is a regression to fix rather than budget; a p95 lag that rises through a stall, or an underrun count that moves, is re-measured in. No ceiling is fixed at zero underruns or zero stalls; a replay row's is exactly what it measured. The replay rows are deterministic, so their ceilings are exactly what they measured, through the end of each recording, with no headroom: any regression fails them, and an improvement should be tightened in. They were last re-measured on 2026-10-22 using version 3 traces with encoded frame durations: AAC-LC uses its fixed 1024 samples, and relay-mic retains Opus packet TOC durations. relay-mic was re-recorded on 2026-11-03 from a real microphone, Chrome 156 on a MacBook Air publishing its built-in microphone through the public relay, recorded by headless Chromium on a Linux host (35 s, minimum RTT 2 ms), which raised its target ceiling from 60 to 80 ms and its silence from 0 to 0.001 as measured. The recording path itself has no added RTT. relay-mic-rtt40 is the 2026-11-03 Chrome row at 40 ms of added RTT on that laptop (180 s, minimum RTT 47 ms), kept as the baseline time stretch improves on; like every replay row, its player_underruns_per_min ceiling is what it measures, not a fixed zero. player_underruns_per_min, the player's own counter that its stats panel shows, is budgeted on the replay rows only; it counts the run-dry re-stalls the tap files as stalled time. On 2026-11-07, after the ordered read stopped delivering groups below its cursor (#8071) and took Rust's hole and age bounds (#8218), relay-bbb was re-measured at 16.1 player underruns/min, 0.025 stalled and 0.02 silence, and relay-mic-rtt40 at 8.2, its discarded audio at 70.9 ms/min and its target p95 at 320 ms rather than 780, since the stale backlog it never played no longer reaches the estimator; neither row lost audio the previous player heard. On 2026-11-08 the player began time-stretching live playback with NetEq's accelerate, deciding on the ring's depth right after each arrival and never expanding, and every replay row was re-measured with it, the browser rows left to the nightly runner. Each row's av_lag_ms, how far the playhead trails the shared clock video plays from, is what it measured: the ring rests one frame ahead of the newest arrival's playout, as the trim it replaces left it, so audio leads the clock by 19 to 23 ms at p50. Against the trim, the A/V lag p50 fell from 0.7, -17.5, -21.5, -18.2 and -15.6 ms to -22.4 on bursty-aac, -22.2 on local-aac, -22.7 on relay-bbb, -19 on relay-mic and -20.1 on relay-mic-rtt40, and its p95 fell on every row, most on relay-bbb (10.5 to -4.2) and relay-mic-rtt40 (-0.9 to -13.4). Underruns stayed where the trim had them, as the latency they would cost is not spent: player_underruns_per_min held at 4.4, 12.1, 16.1 and 6.1 and rose from 8.2 to 10.6 on relay-mic-rtt40, and relay-mic's tap now files 2 short quanta a minute, 2 ms of missing audio, where it filed none. What the stretch buys is the audio the trim dropped: discarded_ms_per_min fell from 87.8 to 0 on relay-bbb and from 127.5 to 100.6 on local-aac, whose remainder drops nothing: two receive stalls 14 and 15.3 s in leave audio 159 and 132 ms behind the clock, and the fast accelerate brings it back within half a second. player_skips_per_min and trimmed_ms_per_min, the ring's own count of re-anchors that dropped buffered audio and the media they dropped, are budgeted beside it, at 0 on every recorded row. relay-mic-stall, added 2026-11-10, is relay-mic with nothing arriving from 15 s to 16.5 s and the held frames then trickling in at twice real time until 18 s; today's player trims 2152.5 ms/min of it and its target p95 rises to 1540 ms. Once durations became whole nanoseconds, relay-bbb's delay, 200 ms plus one 1024-sample AAC frame, rounds up by a fraction of a sample, so its ring holds one more sample (23 µs) and its A/V lag p50 measures -22.6 rather than -22.7. Neither replay calibration changed the browser or native ceilings. bursty-aac is the 2026-10-18 AAC bursty isolated capture, recorded with the observer subscribed at the decoder's 2 s budget, so it holds every group the player saw. The native rows are provisional: measured on a loaded developer host (x86_64 Linux, the private PulseAudio null sink, seed 7, 60 s a row) on 2026-10-17, three runs of every row from the debug build nightly uses plus three more of the AAC rows from a release build, which played alike. Each ceiling is the worst clean run times 1.25, plus one native hiccup sized from the unimpaired Opus row's worst run (20 underruns/min, 50 gap ms/min, 2.2 skips/min, 110 discarded ms/min, 0.01 stalled, 0.005 silence, one 20 ms bucket of target), rounded up to two significant figures. On 2026-11-10, after native playback began reporting its A/V lag (#8361), the native rows' player_skips_per_min, trimmed_ms_per_min, and av_lag_ms p50 and p95, and every key of the native and browser congestion rows, were measured by the same rules from three runs of each on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row), with no void; native audio rests 9 to 95 ms behind its clock at p50. These are provisional until the stretch model's nightly re-measure, which also moves the browser congestion rows onto the runner. On the 2026-10-17 host the null sink's clock ran about 150 ppm slow against the monotonic clock, read within 5 ms at p95 per sample, and its output latency sat between 14 and 28 ms. Five of the 36 AAC rows (none of the 18 Opus rows) voided on the clock: the device stopped pulling for about two seconds mid-row, sometimes after a backend error and a stream restart, which is a native playback defect rather than noise, and the ceilings exclude those runs. Re-measure from the nightly runner's artifacts once it has a week of native rows. On 2026-11-11 every player began following one stretch rule on the shared clock: it slows down below its target, speeds up above it, and skips only past its bound, which in the browser is the subscription's max delay. Every replay row was re-measured with it, the browser and native rows left to the nightly runner. Audio now rests on the clock rather than one frame ahead of it, so the A/V lag p50 rose from about -20 ms to 0.4 ms on relay-bbb, 1.1 on relay-mic, 1.2 on relay-mic-rtt40 and 1.8 on local-aac. Its p95 rose on every row, since audio that falls behind is now caught up by speeding up rather than trimmed: the shared clock steps forward as its reference converges through a tune-in backlog, and again when a target falls, and audio follows at the accelerate's pace, which on the replay's constant tone is one 2.5 ms period a window, slower than on speech. bursty-aac rests 5 ms behind its clock, so its p50 of 10.4 ms is mostly its catch-ups, and relay-mic-stall's p50 of 66.1 ms is the catch-up after its target falls from 1540 to 80 ms, which outlasts half the row. In exchange nothing is trimmed: relay-mic-stall's trimmed_ms_per_min fell from 2152.5 to 0, its skips from 6.1 to 0, and its underruns from 24.2 to 0 a minute, and player_underruns_per_min fell from 12.1 to 4 on local-aac, 16.1 to 0 on relay-bbb, 6.1 to 0 on relay-mic, 10.6 to 2.7 on relay-mic-rtt40 and 6.1 to 4 on relay-mic-stall, holding at 4.4 on bursty-aac. The tap now files 3.3 to 6.1 short quanta a minute on the AAC rows, under 0.3 ms of missing audio, where it filed none. To deliver the audio a stall delays, the container consumer now waits and skips at the subscription's max delay, as native already does, rather than at the delay plus lookahead. relay-mic-rtt40's join backlog arrives reordered, so the consumer now waits on its holes and the estimator observes the frames that fill them, up to 775 ms late, and its target p95 returns from 320 to 780 ms, held for one half-life after the join. Later on 2026-11-11 audio began skipping as far as the shared clock's reference moves earlier, once those moves add up to one stretch window and only as far as it holds audio past its target, rather than speeding through them, and every replay row was re-measured. The tune-in backlog's jumps now land before the graded window, so the A/V lag p95 fell from 453.1 to 2.1 ms on relay-bbb (its p50 from 0.4 to -1.3), 743.7 to 1.9 on relay-mic-rtt40, whose p50 fell from 1.2 to 0.4 and its tap's missing audio rose from 0.6 to 0.9 ms a minute, since a jump past what is decoded skips the rest once it lands,, 335.5 to 39.2 on local-aac and 88.6 to 0.3 on relay-mic, and the p50 from 66.1 to 25.3 on relay-mic-stall, with nothing trimmed in any graded window. The tap's underruns fell from 6.1 to 2 a minute on local-aac and 6.1 to 0 on relay-bbb; local-aac's silence rose from 0.001 to 0.002. Later on 2026-11-11 the container consumer began waiting on a missing group only for the delay plus the lookahead, while still delivering groups that arrived until they age past the max delay, and every replay row was re-measured. relay-mic-rtt40's target p95 fell from 780 back to 320 ms, since the holes in its reordered join backlog no longer delay the frames behind them; its tap's missing audio rose from 0.9 to 1.1 ms a minute and its A/V lag p50 and p95 from 0.4 and 1.9 to 0.7 and 2.4 ms. bursty-aac's silence rose from 0.007 to 0.009 and its A/V lag p95 fell from 162.1 to 156.3 ms, and relay-bbb's lag p50 and p95 moved from -1.3 and 2.1 to -1.2 and 2.2 ms. relay-mic-stall did not move. Also on 2026-11-11 the stretch became a player setting, 2 s by default, and the audio subscription's max delay became the delay plus it rather than the 2 s constant, so the container consumer delivers a late group that much longer before it ages out, and every replay row was re-measured: only relay-mic moved, its A/V lag p50 and p95 from 0.3 to 1.6 ms, under the 2.5 ms pitch period the stretch can move it by. Holding the constant max delay with the new bound reproduces the old values exactly. On 2026-11-12 the replays stopped playing a constant and began playing a seeded speech-like signal (syllables gliding a 90 to 220 Hz fundamental, with pauses held above the silence classifier), since the stretch's pitch search found only its shortest period in a constant, and the browser ring stopped zero-filling a hole of up to a millisecond and two samples between frames, as native stopped writing silence for one: the recorded AAC timestamps land up to 2.5 samples either side of a 1024-sample boundary, and on speech the 2-sample holes this filled ended 119.1 quanta a minute short on bursty-aac. Every replay row was re-measured. The tap's underruns fell to 0 on bursty-aac (from 3.3) and local-aac (from 2), and relay-mic-rtt40's missing audio from 1.1 to 0.3 ms a minute while its player underruns rose from 2.7 to 3.1. The A/V lag p95 fell from 156.3 to 104.8 ms on bursty-aac, 39.2 to 15.1 on local-aac, 2.2 to 1.4 on relay-bbb and 501.8 to 484.9 on relay-mic-stall, whose p50 fell from 25.3 to 1.1, and rose from 1.6 to 3.4 on relay-mic and 2.4 to 3.5 on relay-mic-rtt40. Where audio rests moved within one speech pitch period, which is as close as one stretch seam can steer it: the p50 is 11 ms on bursty-aac (from 10.4), -4.2 on relay-bbb (from -1.2), 1.1 on relay-mic (from 1.6), 0.9 on relay-mic-rtt40 (from 0.7) and 1.7 on local-aac (from 1.8). On 2026-11-13 the stretch default fell from 2 s to 200 ms and audio trailing the clock past it began skipping onto the clock rather than back to the bound, as video skips, with the audio subscription still reaching 2 s past the delay; every replay row was re-measured. A target fall past 200 ms now costs one skip rather than seconds of catch-up, and so does a trail past it, and a rise past it pads silence until the clock arrives rather than slowing down: relay-mic-stall skips the 1.46 s it trails after the outage and its 1540 to 1020 and 1020 to 80 ms falls, and pads the 0.5 s its 60 to 1540 ms rise leaves it ahead, where it slowed down for 4 s, so it trims 4678.2 ms a minute where it trimmed none and its stalled share rose from 0.05 to 0.067 and silence from 0.049 to 0.085. The frames it holds through the outage then trickle in at twice real time, so its low buffer slows it to 154 ms behind the clock, where it had sat ahead of it: its A/V lag p95 fell from 484.9 to 118.6 ms and its p50 rose from 1.1 to 2.2. relay-mic-rtt40 skips its 320 to 80 ms fall, 0.3 skips and 83.3 trimmed ms a minute, its p95 from 3.5 to 2.9 ms. bursty-aac skips a 225 ms trail 6 s in, 1.1 skips and 312.7 trimmed ms a minute, its p95 from 102.7 to 49.2 ms and p50 from 11 to 10. relay-mic trims nothing, but its tune-in's skips land on the clock rather than at the bound, which moves where it rests within a pitch period: its p50 from 1.1 to 2.2 ms and p95 from 3.4 to 4.5. Trimming at a target fall or a trail past 200 ms is the rule, not a regression. Also on 2026-11-13, for the same change, the congestion rows' trimmed_ms_per_min ceilings were raised by the same rule from two runs of each on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row): to 3800 on chromium AAC (worst 2740.3 ms/min), 3700 on chromium Opus (2697.2), 3900 on native AAC (2996.9) and 4600 on native Opus (3532.3), whose player_skips_per_min rose to 15 (9.8). The trim is the outage's audio skipped onto the clock rather than played seconds late: against main on the same host, chromium's A/V lag p95 fell from 1396.8 to 61.6 ms on AAC and 1424.8 to 64.3 on Opus, and native's from 1246.3 to 116.7 and 826.9 to 117.5. These four ceilings are provisional until the nightly re-measures the congestion rows. Also on 2026-11-13, the estimator began counting one outage, the intervals its stalled frames span and the empty ones before them, as one observation, and every replay row was re-measured on top of the snap: only relay-mic-stall moved. Its single stall no longer raises the target, so its target p95 fell from 1540 to 80 ms and the trail the outage leaves is skipped onto the clock at once rather than through the target's falls: 2 skips a minute where it took 6.1, 5122.3 trimmed ms a minute where it trimmed 4678.2, and its A/V lag p95 from 118.6 to 19.7 ms. With 80 ms of buffer rather than 1540, more of the outage plays as silence: its stalled share rose from 0.067 to 0.092 and silence from 0.085 to 0.092. On 2026-11-14 every player began following the clock by its lag alone: a continuous WSOLA time stretch for any content, at a rate proportional to the lag and capped at 10%, which leaves a lag within half a 10 ms hop alone, and slows down while the audio received ahead of the playhead is about to run out, replacing NetEq's buffer-level decision and pitch-period splices. Every replay row was re-measured. Audio now rests within half a hop of the clock rather than within a pitch period: the A/V lag p50 fell from 10 to 0.2 ms on bursty-aac, whose p95 fell from 49.2 to 0.2, from 1.7 to -1.3 on local-aac, 2.2 to 1.8 on relay-mic and 0.9 to 0.6 on relay-mic-rtt40, and moved from -2 to 2.7 on relay-bbb and 2.2 to 3.1 on relay-mic-stall, each within the half hop the rule leaves alone. The p95 rose from 3.6 to 6.2 ms on relay-bbb, 4.5 to 6.9 on relay-mic, 2.9 to 6.9 on relay-mic-rtt40 and 19.7 to 32.1 on relay-mic-stall, and fell from 15.1 to 14.3 on local-aac: a lag under a hop is corrected only once its pace adds up to one. player_underruns_per_min rose from 4.4 to 6.6 on bursty-aac and 4 to 8.1 on local-aac, whose recorded outages outlast what slowing down at 10% bridges, and fell from 3.1 to 1 on relay-mic-rtt40, whose tap's underruns fell from 0.7 to 0.3 a minute; bursty-aac's silence rose from 0.007 to 0.008. Nothing more is trimmed on any row. The browser and native rows were measured on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row) against main the same day and left to the nightly runner: every browser row's A/V lag p50 now sits between 27 and 39 ms, where the fixed-250 control does on that host, where wide and bursty sat up to 58 ms behind, and every native row's within 15 ms of the clock. Also on 2026-11-14 the estimator stopped folding the intervals a subscription's join spans, those holding a frame due before its backlog finished arriving, and every replay row was re-measured: only relay-mic and relay-mic-stall moved. Their first frame arrives slower than the path after it, so their first interval folds nothing and the target leaves the prior 500 ms later, which moves where audio rests within the half hop the rule leaves alone: relay-mic's A/V lag p50 from 1.8 to -0.3 ms and p95 from 6.9 to 9, and relay-mic-stall's p50 from 3.1 to 1.7. relay-mic-rtt40 did not move: the browser consumer closes its join's stale groups below the cursor unread, so the estimator never sees when that backlog ended, and its target p95 holds at 320 ms. Later on 2026-11-14 both container consumers began reporting when each group they close unread arrived, if it is at or below the publisher's largest group as the subscription began, which extends a join still open. The replay's track learns its largest from the first group to land, relay-mic-rtt40's stale tune-in frame, below the backlog it closes, so no replay row moved; with that backlog reported, its target p95 falls from 320 to 120 ms, which a recording carrying the subscribe answer's largest would measure.", + "note": "The browser rows are measured on the nightly runner (ubuntu-24.04-arm, headless Chromium, seed 7, 60 s a row), from the audio-quality-run artifacts of the seven scheduled Nightly runs on main from 2026-10-29 to 2026-11-04 (40885347517, 41065371094, 41204380716, 41313466339, 41483654426, 41656613559, 41852517544), the first after a fresh subscriber stopped receiving the relay's cached history. The six AAC rows other than congestion are measured the same way from the seven runs from 2026-11-05 to 2026-11-11 (42050683248, 42236107954, 42377867523, 42487375545, 42659882487, 42842827718, 43041909115), the first after the TS importer published each PES as one group; the first four predate the browser's time stretch, and the six rows keep the four provisional ceilings set below from 2026-11-09 and 2026-11-10. Each ceiling is the worst of its seven nights times 1.25, plus one runner hiccup, rounded up to two significant figures. The hiccup is sized from the single gaps the 2026-10-08 night put on otherwise clean rows, the fixed-250 control among them: 250 underruns/min, 700 gap ms/min, 3.3 skips/min, 280 dropped ms/min (both sized on the playhead's skip-ahead inference, and now applied to player_skips_per_min and trimmed_ms_per_min), 0.01 stalled, 0.015 silence, and one 20 ms bucket of target. Budgeted this way, every night passes when it is held out of its row's measurement except one, and that row's own ceiling includes it: the 2026-11-03 Opus bursty row raised its target to 300 ms 2.6 s in and held it for one 20 s half-life before settling at 180. Such startup holds, Opus near-zero's 100 ms on 2026-10-30 and 10-31 among them, are budgeted as measured. These ceilings record today's player as the runner hears it, not target audio quality. To re-measure, download the same artifact from the latest scheduled nightlies (`gh run download -n audio-quality-run`), which keep every run, and apply the same rule. On 2026-11-10 the inferred skip_aheads and discarded_ms keys were dropped, since they read a time stretch's catch-up as dropped audio, and every browser row gained player_skips_per_min, trimmed_ms_per_min, and av_lag_ms p50 and p95, by the same rule from the only two scheduled nights on main with today's browser player and its trim counter, 2026-11-09 and 2026-11-10 (42659882487, 42842827718). A lag is signed, so its ceiling is the worst night plus a quarter of its size plus one 20 ms bucket. These four keys are provisional until seven nights with the stretch model re-measure every row. The grade follows the model: audio rests on the shared clock, trails it through a stall, and catches up after. When a change moves a row, a steady-state lag (av_lag_ms p50) that rises, or audio the player trimmed, is a regression to fix rather than budget; a p95 lag that rises through a stall, or an underrun count that moves, is re-measured in. No ceiling is fixed at zero underruns or zero stalls; a replay row's is exactly what it measured. The replay rows are deterministic, so their ceilings are exactly what they measured, through the end of each recording, with no headroom: any regression fails them, and an improvement should be tightened in. They were last re-measured on 2026-10-22 using version 3 traces with encoded frame durations: AAC-LC uses its fixed 1024 samples, and relay-mic retains Opus packet TOC durations. relay-mic was re-recorded on 2026-11-03 from a real microphone, Chrome 156 on a MacBook Air publishing its built-in microphone through the public relay, recorded by headless Chromium on a Linux host (35 s, minimum RTT 2 ms), which raised its target ceiling from 60 to 80 ms and its silence from 0 to 0.001 as measured. The recording path itself has no added RTT. relay-mic-rtt40 is the 2026-11-03 Chrome row at 40 ms of added RTT on that laptop (180 s, minimum RTT 47 ms), kept as the baseline time stretch improves on; like every replay row, its player_underruns_per_min ceiling is what it measures, not a fixed zero. player_underruns_per_min, the player's own counter that its stats panel shows, is budgeted on the replay rows only; it counts the run-dry re-stalls the tap files as stalled time. On 2026-11-07, after the ordered read stopped delivering groups below its cursor (#8071) and took Rust's hole and age bounds (#8218), relay-bbb was re-measured at 16.1 player underruns/min, 0.025 stalled and 0.02 silence, and relay-mic-rtt40 at 8.2, its discarded audio at 70.9 ms/min and its target p95 at 320 ms rather than 780, since the stale backlog it never played no longer reaches the estimator; neither row lost audio the previous player heard. On 2026-11-08 the player began time-stretching live playback with NetEq's accelerate, deciding on the ring's depth right after each arrival and never expanding, and every replay row was re-measured with it, the browser rows left to the nightly runner. Each row's av_lag_ms, how far the playhead trails the shared clock video plays from, is what it measured: the ring rests one frame ahead of the newest arrival's playout, as the trim it replaces left it, so audio leads the clock by 19 to 23 ms at p50. Against the trim, the A/V lag p50 fell from 0.7, -17.5, -21.5, -18.2 and -15.6 ms to -22.4 on bursty-aac, -22.2 on local-aac, -22.7 on relay-bbb, -19 on relay-mic and -20.1 on relay-mic-rtt40, and its p95 fell on every row, most on relay-bbb (10.5 to -4.2) and relay-mic-rtt40 (-0.9 to -13.4). Underruns stayed where the trim had them, as the latency they would cost is not spent: player_underruns_per_min held at 4.4, 12.1, 16.1 and 6.1 and rose from 8.2 to 10.6 on relay-mic-rtt40, and relay-mic's tap now files 2 short quanta a minute, 2 ms of missing audio, where it filed none. What the stretch buys is the audio the trim dropped: discarded_ms_per_min fell from 87.8 to 0 on relay-bbb and from 127.5 to 100.6 on local-aac, whose remainder drops nothing: two receive stalls 14 and 15.3 s in leave audio 159 and 132 ms behind the clock, and the fast accelerate brings it back within half a second. player_skips_per_min and trimmed_ms_per_min, the ring's own count of re-anchors that dropped buffered audio and the media they dropped, are budgeted beside it, at 0 on every recorded row. relay-mic-stall, added 2026-11-10, is relay-mic with nothing arriving from 15 s to 16.5 s and the held frames then trickling in at twice real time until 18 s; today's player trims 2152.5 ms/min of it and its target p95 rises to 1540 ms. Once durations became whole nanoseconds, relay-bbb's delay, 200 ms plus one 1024-sample AAC frame, rounds up by a fraction of a sample, so its ring holds one more sample (23 µs) and its A/V lag p50 measures -22.6 rather than -22.7. Neither replay calibration changed the browser or native ceilings. bursty-aac is the 2026-10-18 AAC bursty isolated capture, recorded with the observer subscribed at the decoder's 2 s budget, so it holds every group the player saw. The native rows are provisional: measured on a loaded developer host (x86_64 Linux, the private PulseAudio null sink, seed 7, 60 s a row) on 2026-10-17, three runs of every row from the debug build nightly uses plus three more of the AAC rows from a release build, which played alike. Each ceiling is the worst clean run times 1.25, plus one native hiccup sized from the unimpaired Opus row's worst run (20 underruns/min, 50 gap ms/min, 2.2 skips/min, 110 discarded ms/min, 0.01 stalled, 0.005 silence, one 20 ms bucket of target), rounded up to two significant figures. On 2026-11-10, after native playback began reporting its A/V lag (#8361), the native rows' player_skips_per_min, trimmed_ms_per_min, and av_lag_ms p50 and p95, and every key of the native and browser congestion rows, were measured by the same rules from three runs of each on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row), with no void; native audio rests 9 to 95 ms behind its clock at p50. These are provisional until the stretch model's nightly re-measure, which also moves the browser congestion rows onto the runner. On the 2026-10-17 host the null sink's clock ran about 150 ppm slow against the monotonic clock, read within 5 ms at p95 per sample, and its output latency sat between 14 and 28 ms. Five of the 36 AAC rows (none of the 18 Opus rows) voided on the clock: the device stopped pulling for about two seconds mid-row, sometimes after a backend error and a stream restart, which is a native playback defect rather than noise, and the ceilings exclude those runs. Re-measure from the nightly runner's artifacts once it has a week of native rows. On 2026-11-11 every player began following one stretch rule on the shared clock: it slows down below its target, speeds up above it, and skips only past its bound, which in the browser is the subscription's max delay. Every replay row was re-measured with it, the browser and native rows left to the nightly runner. Audio now rests on the clock rather than one frame ahead of it, so the A/V lag p50 rose from about -20 ms to 0.4 ms on relay-bbb, 1.1 on relay-mic, 1.2 on relay-mic-rtt40 and 1.8 on local-aac. Its p95 rose on every row, since audio that falls behind is now caught up by speeding up rather than trimmed: the shared clock steps forward as its reference converges through a tune-in backlog, and again when a target falls, and audio follows at the accelerate's pace, which on the replay's constant tone is one 2.5 ms period a window, slower than on speech. bursty-aac rests 5 ms behind its clock, so its p50 of 10.4 ms is mostly its catch-ups, and relay-mic-stall's p50 of 66.1 ms is the catch-up after its target falls from 1540 to 80 ms, which outlasts half the row. In exchange nothing is trimmed: relay-mic-stall's trimmed_ms_per_min fell from 2152.5 to 0, its skips from 6.1 to 0, and its underruns from 24.2 to 0 a minute, and player_underruns_per_min fell from 12.1 to 4 on local-aac, 16.1 to 0 on relay-bbb, 6.1 to 0 on relay-mic, 10.6 to 2.7 on relay-mic-rtt40 and 6.1 to 4 on relay-mic-stall, holding at 4.4 on bursty-aac. The tap now files 3.3 to 6.1 short quanta a minute on the AAC rows, under 0.3 ms of missing audio, where it filed none. To deliver the audio a stall delays, the container consumer now waits and skips at the subscription's max delay, as native already does, rather than at the delay plus lookahead. relay-mic-rtt40's join backlog arrives reordered, so the consumer now waits on its holes and the estimator observes the frames that fill them, up to 775 ms late, and its target p95 returns from 320 to 780 ms, held for one half-life after the join. Later on 2026-11-11 audio began skipping as far as the shared clock's reference moves earlier, once those moves add up to one stretch window and only as far as it holds audio past its target, rather than speeding through them, and every replay row was re-measured. The tune-in backlog's jumps now land before the graded window, so the A/V lag p95 fell from 453.1 to 2.1 ms on relay-bbb (its p50 from 0.4 to -1.3), 743.7 to 1.9 on relay-mic-rtt40, whose p50 fell from 1.2 to 0.4 and its tap's missing audio rose from 0.6 to 0.9 ms a minute, since a jump past what is decoded skips the rest once it lands,, 335.5 to 39.2 on local-aac and 88.6 to 0.3 on relay-mic, and the p50 from 66.1 to 25.3 on relay-mic-stall, with nothing trimmed in any graded window. The tap's underruns fell from 6.1 to 2 a minute on local-aac and 6.1 to 0 on relay-bbb; local-aac's silence rose from 0.001 to 0.002. Later on 2026-11-11 the container consumer began waiting on a missing group only for the delay plus the lookahead, while still delivering groups that arrived until they age past the max delay, and every replay row was re-measured. relay-mic-rtt40's target p95 fell from 780 back to 320 ms, since the holes in its reordered join backlog no longer delay the frames behind them; its tap's missing audio rose from 0.9 to 1.1 ms a minute and its A/V lag p50 and p95 from 0.4 and 1.9 to 0.7 and 2.4 ms. bursty-aac's silence rose from 0.007 to 0.009 and its A/V lag p95 fell from 162.1 to 156.3 ms, and relay-bbb's lag p50 and p95 moved from -1.3 and 2.1 to -1.2 and 2.2 ms. relay-mic-stall did not move. Also on 2026-11-11 the stretch became a player setting, 2 s by default, and the audio subscription's max delay became the delay plus it rather than the 2 s constant, so the container consumer delivers a late group that much longer before it ages out, and every replay row was re-measured: only relay-mic moved, its A/V lag p50 and p95 from 0.3 to 1.6 ms, under the 2.5 ms pitch period the stretch can move it by. Holding the constant max delay with the new bound reproduces the old values exactly. On 2026-11-12 the replays stopped playing a constant and began playing a seeded speech-like signal (syllables gliding a 90 to 220 Hz fundamental, with pauses held above the silence classifier), since the stretch's pitch search found only its shortest period in a constant, and the browser ring stopped zero-filling a hole of up to a millisecond and two samples between frames, as native stopped writing silence for one: the recorded AAC timestamps land up to 2.5 samples either side of a 1024-sample boundary, and on speech the 2-sample holes this filled ended 119.1 quanta a minute short on bursty-aac. Every replay row was re-measured. The tap's underruns fell to 0 on bursty-aac (from 3.3) and local-aac (from 2), and relay-mic-rtt40's missing audio from 1.1 to 0.3 ms a minute while its player underruns rose from 2.7 to 3.1. The A/V lag p95 fell from 156.3 to 104.8 ms on bursty-aac, 39.2 to 15.1 on local-aac, 2.2 to 1.4 on relay-bbb and 501.8 to 484.9 on relay-mic-stall, whose p50 fell from 25.3 to 1.1, and rose from 1.6 to 3.4 on relay-mic and 2.4 to 3.5 on relay-mic-rtt40. Where audio rests moved within one speech pitch period, which is as close as one stretch seam can steer it: the p50 is 11 ms on bursty-aac (from 10.4), -4.2 on relay-bbb (from -1.2), 1.1 on relay-mic (from 1.6), 0.9 on relay-mic-rtt40 (from 0.7) and 1.7 on local-aac (from 1.8). On 2026-11-13 the stretch default fell from 2 s to 200 ms and audio trailing the clock past it began skipping onto the clock rather than back to the bound, as video skips, with the audio subscription still reaching 2 s past the delay; every replay row was re-measured. A target fall past 200 ms now costs one skip rather than seconds of catch-up, and so does a trail past it, and a rise past it pads silence until the clock arrives rather than slowing down: relay-mic-stall skips the 1.46 s it trails after the outage and its 1540 to 1020 and 1020 to 80 ms falls, and pads the 0.5 s its 60 to 1540 ms rise leaves it ahead, where it slowed down for 4 s, so it trims 4678.2 ms a minute where it trimmed none and its stalled share rose from 0.05 to 0.067 and silence from 0.049 to 0.085. The frames it holds through the outage then trickle in at twice real time, so its low buffer slows it to 154 ms behind the clock, where it had sat ahead of it: its A/V lag p95 fell from 484.9 to 118.6 ms and its p50 rose from 1.1 to 2.2. relay-mic-rtt40 skips its 320 to 80 ms fall, 0.3 skips and 83.3 trimmed ms a minute, its p95 from 3.5 to 2.9 ms. bursty-aac skips a 225 ms trail 6 s in, 1.1 skips and 312.7 trimmed ms a minute, its p95 from 102.7 to 49.2 ms and p50 from 11 to 10. relay-mic trims nothing, but its tune-in's skips land on the clock rather than at the bound, which moves where it rests within a pitch period: its p50 from 1.1 to 2.2 ms and p95 from 3.4 to 4.5. Trimming at a target fall or a trail past 200 ms is the rule, not a regression. Also on 2026-11-13, for the same change, the congestion rows' trimmed_ms_per_min ceilings were raised by the same rule from two runs of each on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row): to 3800 on chromium AAC (worst 2740.3 ms/min), 3700 on chromium Opus (2697.2), 3900 on native AAC (2996.9) and 4600 on native Opus (3532.3), whose player_skips_per_min rose to 15 (9.8). The trim is the outage's audio skipped onto the clock rather than played seconds late: against main on the same host, chromium's A/V lag p95 fell from 1396.8 to 61.6 ms on AAC and 1424.8 to 64.3 on Opus, and native's from 1246.3 to 116.7 and 826.9 to 117.5. These four ceilings are provisional until the nightly re-measures the congestion rows. Also on 2026-11-13, the estimator began counting one outage, the intervals its stalled frames span and the empty ones before them, as one observation, and every replay row was re-measured on top of the snap: only relay-mic-stall moved. Its single stall no longer raises the target, so its target p95 fell from 1540 to 80 ms and the trail the outage leaves is skipped onto the clock at once rather than through the target's falls: 2 skips a minute where it took 6.1, 5122.3 trimmed ms a minute where it trimmed 4678.2, and its A/V lag p95 from 118.6 to 19.7 ms. With 80 ms of buffer rather than 1540, more of the outage plays as silence: its stalled share rose from 0.067 to 0.092 and silence from 0.085 to 0.092. On 2026-11-14 every player began following the clock by its lag alone: a continuous WSOLA time stretch for any content, at a rate proportional to the lag and capped at 10%, which leaves a lag within half a 10 ms hop alone, and slows down while the audio received ahead of the playhead is about to run out, replacing NetEq's buffer-level decision and pitch-period splices. Every replay row was re-measured. Audio now rests within half a hop of the clock rather than within a pitch period: the A/V lag p50 fell from 10 to 0.2 ms on bursty-aac, whose p95 fell from 49.2 to 0.2, from 1.7 to -1.3 on local-aac, 2.2 to 1.8 on relay-mic and 0.9 to 0.6 on relay-mic-rtt40, and moved from -2 to 2.7 on relay-bbb and 2.2 to 3.1 on relay-mic-stall, each within the half hop the rule leaves alone. The p95 rose from 3.6 to 6.2 ms on relay-bbb, 4.5 to 6.9 on relay-mic, 2.9 to 6.9 on relay-mic-rtt40 and 19.7 to 32.1 on relay-mic-stall, and fell from 15.1 to 14.3 on local-aac: a lag under a hop is corrected only once its pace adds up to one. player_underruns_per_min rose from 4.4 to 6.6 on bursty-aac and 4 to 8.1 on local-aac, whose recorded outages outlast what slowing down at 10% bridges, and fell from 3.1 to 1 on relay-mic-rtt40, whose tap's underruns fell from 0.7 to 0.3 a minute; bursty-aac's silence rose from 0.007 to 0.008. Nothing more is trimmed on any row. The browser and native rows were measured on a loaded developer host (x86_64 Linux, the debug build, seed 7, 60 s a row) against main the same day and left to the nightly runner: every browser row's A/V lag p50 now sits between 27 and 39 ms, where the fixed-250 control does on that host, where wide and bursty sat up to 58 ms behind, and every native row's within 15 ms of the clock. Also on 2026-11-14 the estimator stopped folding the intervals a subscription's join spans, those holding a frame due before its backlog finished arriving, and every replay row was re-measured: only relay-mic and relay-mic-stall moved. Their first frame arrives slower than the path after it, so their first interval folds nothing and the target leaves the prior 500 ms later, which moves where audio rests within the half hop the rule leaves alone: relay-mic's A/V lag p50 from 1.8 to -0.3 ms and p95 from 6.9 to 9, and relay-mic-stall's p50 from 3.1 to 1.7. relay-mic-rtt40 did not move: the browser consumer closes its join's stale groups below the cursor unread, so the estimator never sees when that backlog ended, and its target p95 held at 320 ms. Later on 2026-11-14 both container consumers began reporting when each group they close unread arrived, if it is at or below the publisher's largest group as the subscription began, which extends a join still open. On 2026-11-15 the traces gained that largest group (version 4), so the replay's consumer learns it as a wire subscriber does, and relay-mic-rtt40 was re-measured: its target p95 fell from 320 to 120 ms, underruns per minute rose from 0.3 to 0.7, player skips per minute fell from 0.3 to 0, trimmed ms per minute from 83 to 0, A/V lag p50 rose from 0.6 to 2.3 ms and p95 from 6.9 to 8.6, the trade of reporting the join's backlog. No other replay row moved.", "rows": [ { "runtime": "chromium", @@ -490,16 +490,16 @@ "codec": "opus", "rate": 48000, "profile": "relay-mic-rtt40", - "player_underruns_per_min": 1, - "player_skips_per_min": 0.3, - "underruns_per_min": 0.3, - "underrun_ms_per_min": 0.2, - "trimmed_ms_per_min": 83, + "player_underruns_per_min": 0.7, + "player_skips_per_min": 0, + "underruns_per_min": 0.7, + "underrun_ms_per_min": 0.8, + "trimmed_ms_per_min": 0, "stalled_share": 0, "silence_share": 0, - "target_ms_p95": 320, - "av_lag_ms_p50": 0.6, - "av_lag_ms_p95": 6.9 + "target_ms_p95": 120, + "av_lag_ms_p50": 2.3, + "av_lag_ms_p95": 8.6 }, { "runtime": "replay", diff --git a/test/audio-quality/clients/js/record.ts b/test/audio-quality/clients/js/record.ts index 376fd06b81e..fc8e09968d7 100644 --- a/test/audio-quality/clients/js/record.ts +++ b/test/audio-quality/clients/js/record.ts @@ -123,15 +123,18 @@ try { arrivals.push(...(await page.evaluate(() => globalThis.recorder.drain()))); // The RTT is read last: the PROBE minimum only falls as the session runs. const rtt = await page.evaluate(() => globalThis.recorder.info()?.rtt ?? null); + const largest = await page.evaluate(() => globalThis.recorder.info()?.largest); + if (largest === undefined) throw new Error("the subscription never reported the publisher's largest group"); const trimmed = arrivals.filter(([at]) => at - start <= durationMs); const round = (x: number, places: number) => Math.round(x * 10 ** places) / 10 ** places; const trace: Trace = { - version: 3, + version: 4, source, description, transport: info.transport, rtt: rtt === null ? null : round(rtt, 1), + largest, config: info.config as Trace["config"], duration: durationMs, arrivals: trimmed.map(([at, timestamp, group, duration]) => [ diff --git a/test/audio-quality/clients/js/replay.ts b/test/audio-quality/clients/js/replay.ts index 4f6e0ef8280..6cb92752b6c 100644 --- a/test/audio-quality/clients/js/replay.ts +++ b/test/audio-quality/clients/js/replay.ts @@ -68,7 +68,7 @@ for (const [name, list, known] of selectors) { const rows: { row: Row; trace: Trace }[] = []; for (const { profile, trace } of traces) { - if (trace.version !== 3) throw new Error(`${profile}: trace version ${trace.version}, expected 3`); + if (trace.version !== 4) throw new Error(`${profile}: trace version ${trace.version}, expected 4`); const codec = traceCodec(trace); if (profiles && !profiles.includes(profile)) continue; if (!codecs.includes(codec)) continue; @@ -141,6 +141,7 @@ for (const { row, trace } of rows) { rate, delay: "auto", config: trace.config as unknown as Catalog.AudioConfig, + largest: trace.largest, duration: trace.duration, }); let last: Reading | undefined; diff --git a/test/audio-quality/clients/js/src/capture.test.ts b/test/audio-quality/clients/js/src/capture.test.ts index c5fe830fa93..7123fa6368c 100644 --- a/test/audio-quality/clients/js/src/capture.test.ts +++ b/test/audio-quality/clients/js/src/capture.test.ts @@ -150,6 +150,51 @@ test("the observer asks for the decoder's delay plus its stretch and warmup, not } }); +test("the observer reads the publisher's largest group as its subscription began, not a later one", async () => { + const origin = new Origin.Producer(); + const publishedRoute = origin.route(Path.from("tone.hang"), { hidden: true }); + const published = publishedRoute.broadcast; + const track = published.createTrack("audio", { timescale: Time.Timescale.MILLI }); + publishedRoute.announce(); + const frame = (group: Group.Producer, micros: number) => + group.writeFrame({ + timestamp: Time.Timestamp.fromMicros(micros), + payload: Legacy.encodeFrame(new Uint8Array([1]), Duration.micros(micros)), + }); + const source = new Broadcast({ origin, name: Path.from("tone.hang"), follow: false, catalogFormat: "manual" }); + const capture = new Capture({ + broadcast: new Signal(source), + track: new Signal("audio"), + config: new Signal({ + codec: "opus", + sampleRate: 48000, + numberOfChannels: 1, + container: { kind: "legacy" }, + } as Catalog.AudioConfig), + delay: new Signal(Duration.millis(250)), + maxDelay: new Signal(Duration.millis(250)), + }); + try { + while (!track.subscription.peek()) await track.subscription.changed(); + const first = track.createGroup(7n); + frame(first, 0); + while (capture.largest() === undefined) await yieldNow(); + // Live groups landing after the answer raise the track's largest, not the join's. + const later = track.createGroup(12n); + frame(later, 20_000); + const seen: Arrival[] = []; + while (seen.length < 2) { + seen.push(...capture.drain()); + await yieldNow(); + } + expect(capture.largest()).toBe(7); + } finally { + await capture.close(); + source.close(); + origin.close(); + } +}); + /** Capture a legacy track, then end its first group with `end` once that group's reader is in flight. */ async function captureGroupEnd(end: (group: Group.Producer) => void): Promise { const origin = new Origin.Producer(); diff --git a/test/audio-quality/clients/js/src/capture.ts b/test/audio-quality/clients/js/src/capture.ts index 2ec38a7c174..e33e2bd1d23 100644 --- a/test/audio-quality/clients/js/src/capture.ts +++ b/test/audio-quality/clients/js/src/capture.ts @@ -4,7 +4,7 @@ import { Duration, Instant, now } from "@moq/clock"; import * as Catalog from "@moq/hang/catalog"; import * as Cmaf from "@moq/hang/cmaf"; import * as Container from "@moq/hang/container"; -import { type Group, Error as NetError } from "@moq/net"; +import { type Group, Error as NetError, type Track } from "@moq/net"; import { Effect, type Getter } from "@moq/signals"; import type { Broadcast } from "@moq/watch"; import { frameDuration, packetDuration } from "../../../../../js/watch/src/audio/config.ts"; @@ -32,6 +32,7 @@ export class Capture { readonly #seen = new Set(); readonly #readers = new Set>(); #failure?: string; + #largest?: number; constructor(input: Input) { this.#signals.run((effect) => { @@ -53,6 +54,15 @@ export class Capture { }, }); const groups = new Set(); + const tuned = Promise.withResolvers(); + effect.spawn(async () => { + try { + // The first subscription's: a resubscribe after a rerun joins a track already flowing. + this.#largest ??= await joined(track, tuned.promise); + } catch (error) { + if (!effect.abort.aborted) this.#failure ??= String(error); + } + }); effect.cleanup(() => { track.close(); for (const group of groups) group.close(); @@ -73,6 +83,7 @@ export class Capture { for (;;) { const group = await effect.race(track.recvGroup()); if (!group) throw new Error("the captured track ended"); + tuned.resolve(); groups.add(group); // Closing a group ends its reader cleanly after the buffered frames, so any // other error here is a real failure, even one surfacing during close. @@ -111,6 +122,11 @@ export class Capture { return this.#arrivals.splice(0); } + /** The publisher's largest group as the subscription began, once known. */ + largest(): number | undefined { + return this.#largest; + } + /** Return the failure that stopped capture, if any. */ error(): string | undefined { return this.#failure; @@ -123,6 +139,18 @@ export class Capture { } } +/** + * The publisher's largest group as `track`'s subscription began, read the way the container consumer + * reads it: once the track's info or `tuned`, its first group, shows the subscription established and + * its start resolved, so a cached group racing the answer, or live groups landing after it, don't set it. + */ +export async function joined(track: Track.Subscriber, tuned: Promise): Promise { + await Promise.race([track.info().catch(() => undefined), tuned]); + await track.start(); + const largest = await track.largest(); + return largest === undefined ? undefined : Number(largest.group); +} + /** Decode the audio container selected by the catalog. */ export function format(config: Catalog.AudioConfig): Container.Format { const container = config.container; diff --git a/test/audio-quality/clients/js/src/probe.ts b/test/audio-quality/clients/js/src/probe.ts index f5632dde895..d768c1bdfc8 100644 --- a/test/audio-quality/clients/js/src/probe.ts +++ b/test/audio-quality/clients/js/src/probe.ts @@ -199,6 +199,7 @@ export function probe(watch: MoqWatch, capture = false): Probe { rate: config?.sampleRate, jitter: config?.jitter, config, + largest: recording?.largest(), contextRate: audio.out.context.peek()?.sampleRate, }; }, diff --git a/test/audio-quality/clients/js/src/recorder.ts b/test/audio-quality/clients/js/src/recorder.ts index 5c7a2d8aa5c..f2116966dfe 100644 --- a/test/audio-quality/clients/js/src/recorder.ts +++ b/test/audio-quality/clients/js/src/recorder.ts @@ -19,7 +19,7 @@ import { Duration } from "@moq/clock"; import * as Catalog from "@moq/hang/catalog"; import * as Moq from "@moq/net"; -import { format, arrivals as readArrivals } from "./capture.ts"; +import { format, joined, arrivals as readArrivals } from "./capture.ts"; import type { Arrival } from "./schema.ts"; /** How far back the relay may serve a group. Long, so a late group is recorded rather than dropped. */ @@ -30,7 +30,9 @@ export type Recorder = { /** Arrivals since the last drain, `at` on this page's `performance.now()`. */ drain(): Arrival[]; /** The session and rendition, once subscribed. */ - info(): { transport: string; rtt: number | null; config: Catalog.AudioConfig } | undefined; + info(): + | { transport: string; rtt: number | null; largest: number | undefined; config: Catalog.AudioConfig } + | undefined; /** The failure that stopped recording, if one did. */ error(): string | undefined; }; @@ -65,6 +67,7 @@ async function record(): Promise { const connection = await Moq.Connection.connect({ url: new URL(required("url")), consume: origin }); let rtt: number | null = null; + let largest: number | undefined; connection.probe.subscribe((probe) => { if (probe.rtt === undefined) return; const sample = probe.rtt.asMillis(); @@ -96,12 +99,22 @@ async function record(): Promise { get rtt() { return rtt; }, + get largest() { + return largest; + }, config, }; + const tuned = Promise.withResolvers(); + joined(track, tuned.promise) + .then((n) => { + largest = n; + }) + .catch(fail); for (;;) { const group = await track.recvGroup(); if (!group) throw new Error("the track ended"); + tuned.resolve(); (async () => { for await (const arrival of readArrivals(group, { decoder, config })) arrivals.push(arrival); })().catch(fail); diff --git a/test/audio-quality/clients/js/src/schema.ts b/test/audio-quality/clients/js/src/schema.ts index a8cbc11b3d7..1f5dde45173 100644 --- a/test/audio-quality/clients/js/src/schema.ts +++ b/test/audio-quality/clients/js/src/schema.ts @@ -316,7 +316,7 @@ export type Arrival = [at: Ms, timestamp: Ms, group: number, duration: Ms]; /** A recorded arrival trace, as checked in under `traces/`. The file name is its profile. */ export type Trace = { /** Bumped when a field's meaning changes. */ - version: 3; + version: 4; /** Where it was recorded: the relay, the broadcast, its publisher, the date. */ source: string; /** The shape it carries, and why it is kept. */ @@ -325,6 +325,11 @@ export type Trace = { transport: string; /** The smallest round trip the connection's PROBE reported, retained as a diagnostic. */ rtt: Ms | null; + /** + * The publisher's largest group as the subscription began, which a wire subscriber learns from the + * publisher's answer: the replay's consumer judges a group it skips unread as join backlog against it. + */ + largest: number; /** The audio rendition, as the catalog advertised it. */ config: { codec: string; @@ -467,6 +472,8 @@ export type Environment = { jitter?: Ms; /** The full audio configuration needed to replay a captured live row. */ config?: Trace["config"]; + /** With `--capture`, the publisher's largest group as the observer's subscription began, which a replayed row needs. */ + largest?: number; /** The AudioContext's rate, Hz, which the render clock counts in. Natively, the output device's. */ contextRate?: number; }; diff --git a/test/audio-quality/traces/bursty-aac.json b/test/audio-quality/traces/bursty-aac.json index 16a57460e25..b165eff8f31 100644 --- a/test/audio-quality/traces/bursty-aac.json +++ b/test/audio-quality/traces/bursty-aac.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "the audio-quality-run artifact of Nightly run 39124936449 on main, 2026-10-18: row chromium-aac-44100-bursty-isolated (ubuntu-24.04-arm, headless Chromium, shaper seed 7), its --capture arrivals", + "version": 4, + "source": "the audio-quality-run artifact of Nightly run 39124936449 on main, 2026-10-18: row chromium-aac-44100-bursty-isolated (ubuntu-24.04-arm, headless Chromium, shaper seed 7), its --capture arrivals. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "This harness's AAC publisher through the shaper's bursty profile. Each seven-frame PES flush is one group per frame, delivered newest first, so most frames arrive after a newer one and the oldest of a flush is the latest. Recorded with the observer subscribed at the decoder's 2 s budget, so every group the player saw is here. No PROBE round trip was recorded: an audio-only page never observes the connection's probe.", "transport": "webtransport", "rtt": null, + "largest": 37665, "config": { "codec": "mp4a.40.2", "container": { diff --git a/test/audio-quality/traces/local-aac.json b/test/audio-quality/traces/local-aac.json index 7bc2595e803..2c154abf629 100644 --- a/test/audio-quality/traces/local-aac.json +++ b/test/audio-quality/traces/local-aac.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "a local moq-relay on 127.0.0.1, ffmpeg's AAC sine over MPEG-TS through moq import ts, recorded 2026-09-29", + "version": 4, + "source": "a local moq-relay on 127.0.0.1, ffmpeg's AAC sine over MPEG-TS through moq import ts, recorded 2026-09-29. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "The shallow control: the audio quality harness's AAC publisher with nothing but loopback between it and the viewer.", "transport": "webtransport", "rtt": 1, + "largest": 42, "config": { "codec": "mp4a.40.2", "container": { diff --git a/test/audio-quality/traces/relay-bbb.json b/test/audio-quality/traces/relay-bbb.json index 5b9a323cf26..103e59c64b0 100644 --- a/test/audio-quality/traces/relay-bbb.json +++ b/test/audio-quality/traces/relay-bbb.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "https://cdn.moq.dev/demo bbb.hang, the public demo's MPEG-TS import, recorded 2026-09-29", + "version": 4, + "source": "https://cdn.moq.dev/demo bbb.hang, the public demo's MPEG-TS import, recorded 2026-09-29. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "The public relay's Big Buck Bunny: one AAC frame per group, flushed in bursts of about seven, the flush-span shape #3477 measured.", "transport": "webtransport", "rtt": null, + "largest": 10065, "config": { "codec": "mp4a.40.2", "container": { diff --git a/test/audio-quality/traces/relay-mic-rtt40.json b/test/audio-quality/traces/relay-mic-rtt40.json index 4f872e6174d..70c5668a0bd 100644 --- a/test/audio-quality/traces/relay-mic-rtt40.json +++ b/test/audio-quality/traces/relay-mic-rtt40.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "https://cdn.moq.dev/anon aq-kixtop-mic-1793728306.hang, in Chrome 156 on a MacBook Air (macOS 26.6.2) capturing its built-in microphone, played by the harness page in a second Chrome 156 profile on the same laptop through a userspace UDP relay adding 20 ms each way, recorded 2026-11-03", + "version": 4, + "source": "https://cdn.moq.dev/anon aq-kixtop-mic-1793728306.hang, in Chrome 156 on a MacBook Air (macOS 26.6.2) capturing its built-in microphone, played by the harness page in a second Chrome 156 profile on the same laptop through a userspace UDP relay adding 20 ms each way, recorded 2026-11-03. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "A real microphone through the public relay with 40 ms of added round trip: the steady-state re-stalls the estimator's 0.95 quantile leaves without time stretch, kept as the baseline time stretch must bring to zero.", "transport": "webtransport", "rtt": 47, + "largest": 8994, "config": { "codec": "opus", "container": { diff --git a/test/audio-quality/traces/relay-mic-stall.json b/test/audio-quality/traces/relay-mic-stall.json index 50f90a86020..438b859f2dc 100644 --- a/test/audio-quality/traces/relay-mic-stall.json +++ b/test/audio-quality/traces/relay-mic-stall.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "relay-mic.json with its arrivals from 15 s rewritten: nothing arrives for 1.5 s, then the held frames arrive oldest first at twice real time until they meet live arrivals 18 s in", + "version": 4, + "source": "relay-mic.json with its arrivals from 15 s rewritten: nothing arrives for 1.5 s, then the held frames arrive oldest first at twice real time until they meet live arrivals 18 s in. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "A path that stalls for longer than the target and then trickles the backlog in late: the audio a player keeps and how long it takes to return to the clock.", "transport": "webtransport", "rtt": 2, + "largest": 32142, "config": { "codec": "opus", "container": { diff --git a/test/audio-quality/traces/relay-mic.json b/test/audio-quality/traces/relay-mic.json index cc84d01f369..bca16ba62ad 100644 --- a/test/audio-quality/traces/relay-mic.json +++ b/test/audio-quality/traces/relay-mic.json @@ -1,9 +1,10 @@ { - "version": 3, - "source": "https://cdn.moq.dev/anon aq-kixtop-mic-1793728306.hang, in Chrome 156 on a MacBook Air (macOS 26.6.2) capturing its built-in microphone, recorded 2026-11-03 by headless Chromium on a Linux host", + "version": 4, + "source": "https://cdn.moq.dev/anon aq-kixtop-mic-1793728306.hang, in Chrome 156 on a MacBook Air (macOS 26.6.2) capturing its built-in microphone, recorded 2026-11-03 by headless Chromium on a Linux host. Version 4 adds largest, the publisher's largest group as the subscription began. This recording predates it, so it is derived (2026-11-15): the newest group whose first frame's media timestamp plus the trace's lowest lag (the smallest at minus timestamp) falls at or before the first arrival, the live edge as the first frame arrived.", "description": "A browser publisher through the public relay: getUserMedia, the capture worklet, and WebCodecs Opus, with the relay's round trip in between.", "transport": "webtransport", "rtt": 2, + "largest": 32142, "config": { "codec": "opus", "container": { From 8cbaf97a6ee250ec16dcd2a31ec91124c7ca61ad Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Sun, 15 Nov 2026 14:59:17 -0800 Subject: [PATCH 3/3] quest: replacement-audio cites the finished replay by its PR Maintainer decision (2026-11-15, /quest-complete): reword the link to the replay-largest quest this PR deletes, so quest check passes on main. Co-Authored-By: Claude Opus 5.5 --- quest/m0/interop-green/replacement-audio.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/quest/m0/interop-green/replacement-audio.md b/quest/m0/interop-green/replacement-audio.md index a1a62c01fa9..d1055bf5be8 100644 --- a/quest/m0/interop-green/replacement-audio.md +++ b/quest/m0/interop-green/replacement-audio.md @@ -29,7 +29,7 @@ Open: where those 600 ms go. The decoder resets the ring on a changed instance whether the ring waits for audio at the clock behind a backlog it must drop (then a join at the live edge is the fix), or the starved page just decodes late (then the bound needs a load term derived from the host, not a constant). Trace the ring's join and un-stall in the worklet on a -failing run; the audio-playout epic's [largest replay](/quest/m2/audio-playout/replay-largest.md) -may overlap. +failing run; the largest-group replay at tune-in +([moq-dev/moq#8844](https://github.com/moq-dev/moq/pull/8844)) may overlap. Public API: none. Wire: none.