MOQ Wiki

MOQ Interop Runner

42 min read

2026-06-07 176/38/137/1; settles after the June-5 drop; 20-day cadence (new longest streak) — 2026-06-07 00:49:07 UTC report: 176 / 38 / 137 / 1 (total / pass / fail / skip; 1 timeout). +1 pass vs June 6 (37 → 38, 21.0% → 21.6%). The 3-day sequence since June 4 is 54 → 41 → 37 → 38 — a −16 cliff into a 37-38 floor, the lowest pass band since the May 19 cadence recovery. Total holds at 176 (one cell dropped from 177 between June 4 and June 6). Version breakdown unchanged: 0 at target · 0 ahead · 176 behind — still no impl registered at draft-18. 20 consecutive daily reports (May 19-Jun 7) = new longest cadence streak the wiki has tracked. Registered versions (June 7 report): aiomoqt 14/16; imquic 16/17; libquicr 14; moq-dev-js 14/16/17; moq-dev-rs 14/16/17; moq-rs 14; moq-rs-draft-16 16; moqlivemock 14/16; moqtail 16; moqx 14/16; moxygen 14/16; quiche-moq 16; xquic 14 — same set as June 4 (no impl moved to draft-18). Carry-forward: the matrix is now ~21% pass against a draft-18 target that no registered impl advertises; the floor is the cost of every cell being a cross-version pairing. The standing levers are unchanged — (1) an impl registering draft-18 (Nokia’s announced v17/v18 relay being the obvious first candidate, Yu You June 2), and (2) Mike English’s promised post-London test-case expansion + streaming-format-level interop (Issue #32). London interim 4 days away (hackathon/interop June 9-10; manual cross-version interop there will exercise 14+16+17+18 beyond the automated runner).

2026-06-06 176/37/138/1; −4 pass; June-5 drop continues — 2026-06-06 00:45:30 UTC report: 176 / 37 / 138 / 1 (total / pass / fail / skip). −4 pass vs June 5 (41 → 37, 23.2% → 21.0%, −2.2pp); total slips 177 → 176 (one cell dropped). 19-day cadence streak. Marks the low point of the post-June-4 slide (38 the next day). Version breakdown still 0 at target · 0 ahead · 176 behind. Plausible attribution: the moq-dev/moq June 4-6 main-branch churn (moq-net Track→TrackInfo reshape PR #1631 +1331/−1217, subscription/N-fetch split PR #1634, moq-gst pad-pump refactor PR #1633) re-broke moq-dev-rs/js cross-version cells that were marginal on June 4 — same wire-level-noise-sensitivity pattern observed May 22-24 and May 28, but deeper here because the moq-net Track/TrackConsumer API was reshaped, not just codecs/CI.

2026-06-05 177/41/135/1; −13 pass; sharp drop from the June-4 baseline — 2026-06-05 00:48:14 UTC report: 177 / 41 / 135 / 1 (total / pass / fail / skip). −13 pass vs June 4 (54 → 41, 30.5% → 23.2%, −7.3pp) — largest single-day pass-count drop the wiki has tracked since cadence recovery May 19, breaking the 50-57 band the matrix had held since May 31. Matrix shape steady at 177; version breakdown unchanged 0 at target · 0 ahead · 177 behind. The −13 is not a target/shape change (target was already draft-18 since June 1, shape unchanged since PR #71 May 25) — it is genuine cross-version regression, most plausibly tied to moq-dev/moq’s June 4-5 main-branch reshape (PR #1631 reshape Track into TrackInfo + async TrackConsumer handle, +1331/−1217, merged June 5 02:24 UTC just before the report cut). When the highest-touch impl reshapes its core track API mid-day, the matrix (running against each impl’s main) registers the breakage on the next cycle. 18-day cadence streak. Carry-forward: watch whether June 6-7 recover (they don’t fully — floor at 37-38) or whether the reshape settles into a new baseline before London.

2026-06-04 177/54/122/0; −3 pass; 17-day cadence (new longest streak the wiki has tracked); third draft-18-target run — 2026-06-04 00:55:39 UTC report: 177 / 54 / 122 / 0 (total / pass / fail / skip). −3 pass vs June 3 (57 → 54, pass rate 32.2% → 30.5%, −1.7pp); skip drops 1 → 0. Matrix shape steady at 177 post-PR-71 expansion + post-PR-68 target shift. 17 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30/31/Jun 1/Jun 2/Jun 3/Jun 4) — new longest cadence streak the wiki has tracked. Rolling 5-day band 51-57, recent trajectory 54 → 51 → 57 → 54 = net flat over the last 4 days (52 → 54 → 51 → 57 → 54). The −3 reverses Jun 3’s +6 jump symmetrically, leaving the 4-day window centered at 54. Third draft-18-target run since PR #68 merged June 1 05:21 UTC. Plausible attribution for the −3 oscillation: (a) moq-dev/moq’s June 3-4 cycle introduced new test surface (PR #1611 apt keyring rename, PR #1619 moq-gst publish, PR #1621 moq-net per-control-stream task, PR #1622 release v0.2.4) that the matrix harness re-categorized; (b) cross-version test cells that passed marginal on June 3 fell back on June 4 with the moq-dev/moq main-branch ~12-merge wave. Same wire-level-noise-sensitivity pattern observed May 22-24 and May 28. Version breakdown still 0 at target · 0 ahead · 177 behind — NO impl has updated its runner registration to advertise draft-18 despite 3 impls (moq-dev/moq, mondain/moqxr, meetecho/imquic) having draft-18 main-branch code; remains the matrix’s unresolved structural gap. Slack signal June 2 09:18 UTC by Yu You (Nokia): “we are in the process to host a remote relay for the interop. we have the support of v17 and v18 (basic ones with current interop test cases) for testing.” — Nokia signals a v17+v18 relay endpoint as pre-London infrastructure addition; if registered with the runner this would be the first impl-registered v18 entry, finally closing the 0-at-target gap. London hackathon 5 days away. Carry-forward: with pass-count oscillating in a 51-57 band centered at 54, and no impl-registration change, the matrix is now in steady-state at ~30% pass rate; the only remaining structural lever is impl registration updates (Nokia’s announced v18 relay being the first candidate). London-week manual interop will exercise (14, 16, 17, 18) × N-impls per the June 1-2 Slack thread (Martin Duke → afrind + Mike English).

2026-06-03 177/57/119/1; +6 pass; new May+June high since cadence recovery; 16-day cadence (new longest); 32.2% pass rate (first 32%+ since cadence recovery May 19) — 2026-06-03 00:56:06 UTC report: 177 / 57 / 119 / 1 (total / pass / fail / skip). +6 pass vs June 2 (51 → 57, pass rate 28.8% → 32.2%, +3.4pp) — largest single-day pass-count jump since the May 19 cadence recovery PR #69 per-test-timeout fix landed. Matrix shape steady at 177 post-PR-71 expansion + post-PR-68 target shift. 16 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30/31/Jun 1/Jun 2/Jun 3) — new longest cadence streak the wiki has tracked. Rolling 5-day band 51-57, trajectory 52 → 54 → 51 → 57 = 3-day net positive (+5 over 3 days). First 32%+ pass rate since cadence recovery. Second draft-18-target run following PR #68 merge June 1 05:21 UTC. Plausibly attributable causes for +6 jump without registry/PR change: (a) the moq-dev/moq June 2-3 ~14-merge cycle (PR #1604 /health endpoint + PR #1606 ergonomic subscribe(None) + PR #1593 AAC esds fMP4 + PR #1595 Frame Start in FETCH) reduced test failure modes; (b) one or more impls now match draft-18 wire behavior even without explicit registry update; (c) matrix harness’s draft-18 expectations adjusted between June 2 and June 3 runs (no observable runner-side PR). Version breakdown still 0 at target · 0 ahead · 177 behind — NO impl has updated its runner registration to advertise draft-18 despite 3 impls having draft-18 main-branch code; the +6 pass rebound therefore comes from better cross-version compatibility behavior, not registration-side drift closure. London hackathon 6 days away. Carry-forward: with pass count at 57/177 (32.2%) the 3-day uptick since the target shift is fully recovering the −3 cost of target recategorization (May 31 52 → Jun 1 54 → Jun 2 51 → Jun 3 57 = net +5 over 3 days post-shift). The remaining matrix-shape lever is each impl updating its runner registration to advertise draft-18; 3 impls have the code (moq-dev/moq, mondain/moqxr, meetecho/imquic), none have updated registration yet. London-week interop will exercise (14, 16, 17, 18) × N-impls manually beyond the automated runner per the June 1-2 Slack thread (Martin Duke → afrind + Mike English).

2026-06-02 177/51/125/1; first run at new draft-18 target; −3 pass; version breakdown 0 at · 0 ahead · 177 behind reveals impl-registration drift; 15-day cadence (new longest) — 2026-06-02 00:50:22 UTC report: 177 / 51 / 125 / 1 (total / pass / fail / skip). First run at the new draft-18 target after PR #68 merge June 1 05:21:25 UTC. −3 pass vs June 1 (54 → 51, pass rate 30.5% → 28.8%, −1.7pp); skip increments 0 → 1. Matrix shape steady at 177 post-PR-71 expansion. 15 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30/31/Jun 1/Jun 2) — extends the longest cadence streak the wiki has tracked by one more day. Version breakdown shifts to 0 at target · 0 ahead · 177 behind — NO impl is registered with the runner as advertising draft-18, even though moq, moqxr, and imquic have draft-18 code in main. Registered version coverage by impl (from June 2 report): moq-rs draft-14; moq-rs-draft-16 draft-16; moxygen draft-14/16; moqx draft-14/16; moq-dev-rs draft-14/16/17; moq-dev-js draft-14/16/17; moqlivemock draft-14/16; xquic draft-14; aiomoqt draft-14/16; quiche-moq draft-16; imquic draft-16/17; libquicr draft-14; moqtail draft-14. Drift between runner-registered-version and impl-main-branch-version is now the visible structural gap. Same-day Slack response (Martin Duke June 1 22:43 UTC → Alan Frindell + Mike English June 2 02:00 UTC): London-week interop will exercise 14+16+partial-18 manually, the matrix is now structurally ahead of every impl — gives the manual London interop work latitude to exercise multi-version freely. −3 pass is consistent with target bump: previously-passing draft-16↔draft-16 cells likely still pass; previously-passing draft-17↔draft-17 cells either remain or shifted to behind-target. The −3 is the cost of recategorization, not regression. London hackathon 7 days away. Carry-forward: with PR #68 exhausting matrix-shape levers and impl-registration drift now the gap, the only remaining structural lever is each impl updating its runner registration to advertise draft-18 (3 impls have the code). Until those registrations update, the matrix tests cross-version pairings against an empty at-target set.

2026-06-01 177/54/122/0; +2 pass; first 30%+ pass rate since cadence recovery; 14-day cadence; 5-day monotonic uptick; PR #68 MERGED — draft-18 target bump finally lands 14 days after OPEN — 2026-06-01 00:49:16 UTC report: 177 / 54 / 122 / 0 (total / pass / fail / skip). +2 pass vs May 31 (52 → 54, pass rate 29.4% → 30.5%, +1.1pp). First 30%+ pass rate since cadence recovery May 19. Matrix shape steady at 177 post-PR-71 expansion. 14 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30/31/Jun 1) — extends the longest cadence streak the wiki has tracked by one more day. Rolling 5-day band 50-54, recent trajectory 50 → 52 → 54; counting from May 28 the 5-day monotonic uptick continues (46 → 49 → 50 → 52 → 54 = +8 pass over 5 days), new May+June high since cadence recovery. MAJOR EVENT: PR #68 “Update interop target to draft-18” MERGED June 1 05:21:25 UTC by englishm-cloudflare — 14 days after OPEN (May 18 21:01 UTC), the first interop-target version bump since matrix inception and first matrix-shape change since PR #71 May 25 (moqx docker adapter). The June 1 00:49 UTC report still shows draft-16 as target (PR merged ~4.5h after report cut); June 2 will be the first run targeting draft-18. Three implementations have been on draft-18 main 2-3 weeks (moq PR #1418 May 18, moqxr May 19-20, imquic May 20); the structural gap (matrix-on-draft-16 vs three-impls-on-draft-18) closes precisely 8 days before London. Under the draft-16 target, version breakdown was 97 at target · 8 ahead · 72 behind; under draft-18 the at-target/ahead/behind decomposition will shift (impls on draft-18 become “at target”, draft-17 becomes “behind”, draft-16 becomes “behind”). Whether this nets pass-count up or down depends on whether the matrix harness counts cross-version pairings as expected-fail or genuine-fail. London hackathon 8 days away. No further structural levers remain — only individual-impl draft-18-conformance fixes can move pass counts now.

2026-05-31 177/52/124/0; +2 pass; 13-day cadence; new May high since cadence recovery; 4-day monotonic uptick — 2026-05-31 00:47:05 UTC report: 177 / 52 / 124 / 0 (total / pass / fail / skip). +2 pass vs May 30 (50 → 52, pass rate 28.2% → 29.4%, +1.2pp). Matrix shape steady at 177 post-PR-71 expansion. 13 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30/31) — extends the longest cadence streak the wiki has tracked by one more day. Rolling 5-day band 46-52, recent trajectory 46 → 49 → 50 → 52 = 4-day monotonic uptick continues and the matrix sets a new May high since cadence recovery (last 52+ was pre-May 14-18 outage). Plausibly attributable to moq-dev/moq’s May 30-31 polyglot-split-day ~14-PR wave (Swift/Kotlin/Go independent FFI splits + moq-rs Python FFI parity gaps + IETF stats path fix + viewer-count atomic refactor + cachix declaration + libiconv scrub) — each merge tweaks the wire/test surface slightly. The PR #1560 “wire session stats into the IETF protocol path” is particularly notable: previously IETF connections silently contributed zero to dashboard/billing counters, so any matrix harness sniffing dashboard signals would have seen IETF cells as “no data” which the harness may now read differently. Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind. London hackathon 9 days away. No new matrix-shape change (PR #68 still gated). With pass count now in 50-52 band with 4-day monotonic uptick and matrix shape untouched since May 25, the draft-18 target bump or PR 71-style adapter additions remain the only structural levers before London.

2026-05-30 177/50/126/0; +1 pass; 12-day cadence; first 50+ since cadence recovery May 19 — 2026-05-30 00:43:27 UTC report: 177 / 50 / 126 / 0 (total / pass / fail / skip). +1 pass vs May 29 (49 → 50, pass rate 27.7% → 28.2%, +0.5pp). Matrix shape steady at 177 post-PR-71 expansion. 12 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29/30) — extends the longest cadence streak the wiki has tracked by one more day. Rolling 5-day band 46-50; recent trajectory 48 → 46 → 49 → 50 = 3-day monotonic uptick continues and the matrix breaks above 50 pass for the first time since cadence recovery May 19. Plausibly attributable to moq-dev/moq’s continued May 29-30 ~15-PR refactor wave (#1528 moq-rtc landed in main; #1542 headless-browser smoke phase 2; #1544 libmoq auto-reconnect; #1546 libmoq terminal-callback lifetime; #1547 kio rename; #1548 stats liveness retention) — each merge tweaks the wire/test surface slightly, the matrix harness picks up new edge passes. Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind. London hackathon 10 days away. No new matrix-shape change (PR #68 still gated). With the pass count now in the 49-50 band and the matrix shape untouched since May 25, all remaining structural improvement before London depends on PR #68’s draft-18 target bump or PR 71-style adapter additions.

2026-05-29 177/49/127/0; +3 pass; 11-day cadence (longest cadence streak the wiki has tracked) — 2026-05-29 00:47:39 UTC report: 177 / 49 / 127 / 0 (total / pass / fail / skip). +3 pass vs May 28 (46 → 49, pass rate 26.0% → 27.7%, +1.7pp). Pass rate above 27% for first time since cadence recovery May 19. Matrix shape steady at 177 post-PR-71 expansion. 11 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28/29) — longest cadence streak the wiki has tracked, extends the May 28 first-double-digit-cadence-streak by one day. Rolling 5-day band 46-49; recent trajectory 48 → 46 → 49 = matrix rebounds past May 27 peak of 48 to a new May high. The +3 is plausibly attributable to moq-dev/moq’s massive May 28-29 ~25-PR refactor wave shaking some test combinations stable: PR #1487 moq-mux catalog filter/Annex-B (+1766/−109, 19 files) + PR #1514 linger upstream subscriptions + PR #1473 timescale refactor + PR #1517 gzipped stats broadcast (+1092/−718) touch a lot of code, and the matrix harness may have picked up new pass combinations previously borderline. Same wire-level-noise-sensitivity-and-recovery pattern as May 22-24 and May 25-27. Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind. London hackathon 11 days away. Same-day downstream signal: Mike English Slack May 27 16:01 UTC asks for streaming-format-level automated interop (Issue #32) — currently matrix tests wire-protocol correctness only, not MSF/CMSF/LOC media-format-level interop; Mike signals this is desired for London but not yet built. PR #68’s draft-18 target bump remains the only outstanding matrix-shape lever before London.

2026-05-28 177/46/130/0; −2 pass; 10-day cadence (first double-digit cadence streak the wiki has tracked) — 2026-05-28 00:41:34 UTC report: 177 / 46 / 130 / 0 (total / pass / fail / skip). −2 pass vs May 27 (48 → 46, pass rate 27.1% → 26.0%, −1.1pp). Matrix shape steady at 177 post-PR-71 expansion. 10 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27/28) — first double-digit cadence streak, exceeding the prior 9-day record. Rolling 5-day band 42-48; recent trajectory 47 → 48 → 46 = the monotonic 2-day uptick from May 25-27 is broken with the third-day −2. The −2 is plausibly attributable to moq-dev/moq’s May 27 refactor wave (PR #1519 multi-tool lint touched 973 LOC + 656 LOC across flake.nix and CI; PR #1515 moq-mux seek(sequence) changed importer trait surfaces; PR #1518 added Lite05Wip version variant that the matrix harness may have noticed) — same wire-level-noise-sensitivity pattern observed May 22-24 and again now. Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind. London hackathon 12 days away. The pass-count is now in a 25-27% band that ~±2 daily refactors swing within; the only structural lever to break out remains PR #68’s draft-18 target bump, which is now the only outstanding matrix-shape event before London.

2026-05-27 177/48/128/0; +1 pass; 9-day cadence (longest sustained streak since May 14-18 outage) — 2026-05-27 00:44:59 UTC report: 177 / 48 / 128 / 0 (total / pass / fail / skip). +1 pass vs May 26 (47 → 48, pass rate 26.6% → 27.1%, +0.5pp). Matrix shape steady at 177 post-PR-71 expansion. 9 consecutive days of daily reports (May 19/20/21/22/23/24/25/26/27) — longest sustained streak since the May 14-18 outage. Rolling 5-day trajectory: 42 → 47 → 48 (last 3 days monotonically up; the matrix is still working off the cumulative effect of PR #71 docker adapter + individual-impl fixes for the new moqx-as-relay cells). Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind. London hackathon 13 days away. The draft-revision lag (matrix on draft-16 vs three impls on draft-18) remains the only outstanding matrix-shape item; PR #68’s merger will be the next discrete event re-shaping the matrix.

2026-05-26 177/47/129/0; first observed matrix-shape change from PR #71; +5 pass / +9 total; 8-day cadence — 2026-05-26 00:43:38 UTC report: 177 / 47 / 129 / 0 (total / pass / fail / skip). Total tests 168 → 177 (+9) — the matrix-shape change from PR #71 (moqx docker adapter, merged May 25 03:18 UTC, after May 25’s report cut) first appears. Pass count 42 → 47 (+5), partial recovery from the May 24-25 −1/−3 sequence. Pass rate 25.0% → 26.6%. Smaller-than-predicted matrix expansion: Giovanni Marzot’s local validation table predicted ~+75 passes with full moqx-as-relay docker adoption; the actual matrix added +9 tests / +5 passes, not +75. Plausible explanations: PR #71 added moqx-as-relay rows against a subset of clients (not all 162-cell columns), or the docker adapter is producing failures the local validation didn’t see. Carry-forward: the gap between gmarzot’s local-validation table and the runner’s observed counts will need a runner-side post-mortem. 8 consecutive days of daily reports (May 19/20/21/22/23/24/25/26) — longest streak since cadence recovery May 19. Target still draft-16 (PR #68 still OPEN since May 18, no new commits since May 19; three implementations on draft-18 main: moq-dev/moq, mondain/moqxr, meetecho/imquic). Version breakdown of 177 tests: 97 at target (draft-16) · 8 ahead · 72 behind.

2026-05-25 168/42/125/0; 7-day cadence; second consecutive regression; moqx docker adapter merged just after report cut — 2026-05-25 00:45:06 UTC report: 168 / 42 / 125 / 0 (total / pass / fail / skip). −3 pass vs May 24 (45 → 42, pass rate 26.8% → 25.0%, −1.8pp). Two consecutive day-over-day regressions (May 24 −1, May 25 −3) following the May 23 +4 recovery; rolling 5-day band tightens to 42–46 pass (trajectory downward 46 → 45 → 42 over 3 days). 7 consecutive days of daily cadence holding (May 19/20/21/22/23/24/25) — extends the post-May-14-18-outage streak. Target still draft-16 (PR #68 OPEN since May 18, no new commits since May 19). Plausible attribution: moq-dev/moq’s continuing main-branch refactor (moq-codec → moq-video rename in PR #1484, moq-clock removal in PR #1494, moq-lite stub removal in PR #1492, moq-mux follow-ons). First matrix-shape change since cadence recovery merged just after report cut: PR #71 MERGED May 25 03:18 UTC by Giovanni Marzot — “moqx: add relay docker adapter, update remote URLs, register publisher” (+52/−2). New adapters/moqx/Dockerfile.relay wraps ghcr.io/openmoq/moqx:latest with the runner’s /certs convention + UDP port 4443; adds roles.relay.docker to the moqx implementation entry; registers publisher role. Expected May 26 matrix impact: moqx-as-relay column moves from 13/18 (no docker, remote only) to ~75+/162 with full docker bringup against moqx + moxygen + moq-rs-* + aiomoqt + moqlivemock clients (all 6/6 in gmarzot’s local validation). Required upstream prereq openmoq/moqx#319 added MOQX_ENDPOINT env var to the moqx docker entrypoint. First post-May-18 matrix-shape change — answers openmoq/moqx#316 (“why does moqx only run 12/12 vs moq-rs’s 18/18?”). The draft-revision lag (matrix on draft-16 vs three+ impls on draft-18) remains the structural gap PR #71 does not address.

2026-05-24 168/45/122/0; 6-day cadence; skip drops 1→0 — 2026-05-24 00:43:56 UTC report: 168 / 45 / 122 / 0 (total / pass / fail / skip). −1 pass vs May 23 (46 → 45, pass rate 27.4% → 26.8%, −0.6pp), skip count drops 1 → 0 (previously-skipped test now runs, contributing to the +1 fail delta). 6 consecutive days of daily cadence holding (May 19/20/21/22/23/24). Target still draft-16 (PR #68 still OPEN). Rolling 5-day picture: 35 → 38 → 46 → 42 → 46 → 45 — matrix has stabilised in a 42–46 pass band. The −1 today is plausibly attributable to moq-dev/moq’s PR #1452 moq-mux restructure (file paths changed for codec/container modules; #1474 follow-on fixes gitignore for moved fmp4 test fixtures). 6-day cadence is the longest consecutive-cadence stretch since the May 14-18 5-day outage.

2026-05-23 recovers to 168/46/121/1; 5-day cadence intact — 2026-05-23 00:42:56 UTC report: 168 / 46 / 121 / 1 (total / pass / fail / skip). +4 pass vs May 22 (42 → 46), pass rate 25.0% → 27.4% (+2.4pp) — fully recovers from May 22’s regression, restoring to May 21 level. 5 consecutive days of daily cadence holding (May 19–23). Skip count returns to 1. The May 22 regression hypothesis (“moq-dev/moq’s May 21 PR cluster broke matrix tests”) appears confirmed: with kixelated’s May 22 merge wave being largely non-wire-level (CMSF/Hang unified pipeline cleanup, LOC frame format, MKV/WebM I/O, stats-via-broadcast, Opus encoder kinds, reconnect timeout) the matrix re-stabilises. Pattern observed: moq-dev/moq main-branch matrix is noise-sensitive to wire-level refactors but quickly self-corrects when downstream PRs are pipeline/codec/library changes. The +4/−4 oscillation over two days shows the matrix registers both regressions and recoveries within one daily cycle. Target still draft-16 (PR #68 still OPEN; no commits since May 19 merge).

2026-05-22 first regression in cadence — 2026-05-22 00:43:51 UTC report: 168 / 42 / 125 / 0 (total / pass / fail / skip). −4 pass vs May 21 (46 → 42), pass rate 27.4% → 25.0% (−2.4pp) — first regression since cadence recovery May 19. 4 consecutive days of daily cadence holding. Skip count drops from 1 → 0 — the one previously-skipped test is now executing and contributes to the +4 fail delta. Hypothesis: May 21’s record-tying moq-dev/moq PR cluster (7 merges including infrastructure-shaping PR #1432 FFI wrappers / PR #1433 origin-consumer refactor / PR #1435 cargo update + PR #1432 Swift+Kotlin wrappers landing May 22 00:38 UTC) likely re-broke some moq-dev-rs/moq-dev-js test combinations stabilised May 19-20. This is the cost of running the matrix against main-of-each-impl — kixelated’s high-velocity main-branch refactoring directly correlates with day-over-day pass-count instability. Target still draft-16 (PR #68 still OPEN).

2026-05-21 daily cadence at Day 3 — 2026-05-21 00:45:51 UTC report: 168 / 46 / 121 / 1 (total / pass / fail / skip). +8 pass-count vs May 20 (38 → 46), pass rate 27.4% (vs 22.6% May 20, +4.8pp — largest single-day pass-rate jump since cadence resumed). 3 consecutive daily reports (May 19 / May 20 / May 21) confirm the per-test-timeout fix structurally stable. Matrix at 13 implementations (aiomoqt, moq-dev-js, moq-dev-rs, moq-rs, moq-rs-draft-16, moqlivemock, moqx, moxygen, xquic, imquic, libquicr, quiche-moq, moqtail); target still draft-16 (PR #68 still OPEN). The pass-rate climb 18.1% → 20.8% → 22.6% → 27.4% over 4 reports (May 13/19/20/21) shows the “low-hanging fruit” Mike English flagged is being addressed implementation-side, even without matrix expansion or target bump. Structural gap widens: imquic merged draft-18 to main May 20 09:25 UTC, joining moq (PR #1418 May 18) and mondain/moqxr (8 commits May 19-20) — 3 implementations now on draft-18 while the matrix targets draft-16.

2026-05-20 daily cadence holding — second consecutive daily report. 2026-05-20 00:46:03 UTC report: 168 / 38 / 129 / 1 (total / pass / fail / skip). +3 pass-count vs May 19 (35 → 38), pass rate 22.6% (vs 20.8% May 19, +1.8pp). Matrix unchanged at 12 implementations; target still draft-16 (PR #68 still OPEN — Mike English working through CI issues, no merge yet). The 2-consecutive-day cadence is the structural confirmation that the May 19 PR #69 per-test-timeout fix worked.

2026-05-19 cadence resumes — first new report in 5 days. 2026-05-19 01:36:37 UTC report: 168 / 35 / 132 (total / pass / fail). Matrix expanded from 105 → 168 tests (+60%) — the May 13 4-PR registry-expansion (mlmtest, aiomoqt, moqx-client, Nokia-via-Docker) is now exercised end-to-end for the first time. The root cause of the May 14–18 silence was diagnosed by mike-english (#moq-interop-runner May 18 21:19 UTC) as a CI-timeout from the combinatorial explosion (test suite hung 6+ hours); PR #69 MERGED May 19 01:35 UTC adds per-test timeouts to prevent hanging tests blocking CI. Status restored to current. PR #68 (Update interop target to draft-18) OPEN May 18 21:01 UTC — staged but not yet merged at report time, so the matrix is still targeting draft-16 vs moq’s shipped draft-18. Note the matrix moq-rs / moq-rs-draft-16 split — cloudflare/moq-rs is now treated as two distinct implementations at different draft revisions. Pass rate 20.8% (vs 18.1% on May 13) is at early-baseline territory: the 4 newly-merged impls are running their first end-to-end validation. Mike English May 19 02:21 UTC: “Probably a lot of low hanging fruit to get some of these to green.”

2026-05-18 status: The interop runner has now missed 5 consecutive daily cadences (May 14, 15, 16, 17, 18). No run has published since May 13 00:41:38 UTC (19 / 72 / 14). Status remains “unreliable”. New structural gap: moq shipped draft-18 May 18 05:08 UTC (first open-source implementation, PR #1418) — the matrix is now structurally one revision behind its highest-touch implementation. With London hackathon 22 days out, the matrix shows nothing of post-May-13 implementation evolution, and the moq-dev-rs / moq-dev-js images now compile against a target the matrix can’t exercise.

2026-05-17 status downgrade: The interop runner has missed 4 consecutive daily cadences (May 14, 15, 16, 17). No run has published since May 13 00:41:38 UTC (19 / 72 / 14). The 4-PR registry expansion merged by mike-english May 13 17:23–17:25 UTC (mlmtest, moqx-client, aiomoqt, Nokia-via-Docker) is uncommitted to a CI run, so the 4 newly registered roles have had zero validation runs under the matrix. Most likely cause is operator bandwidth (Mike has been focused on London interim logistics + cdn-provisioning + relay-dos drafts). Status downgraded from “current” to “unreliable” until a successful run resumes.

Standardized cross-implementation test framework for MOQ protocol interoperability testing.

Overview

The interop runner automates testing between MOQ implementations, publishing results as a matrix showing pass/fail status for each implementation pair.

GitHub: englishm/moq-interop-runner Results: englishm.github.io/moq-interop-runner

Registered Implementations (as of May 2026)

  1. moq-dev-js - moq JavaScript/Hang player (luke-curley)
  2. moq-dev-rs - moq Rust (luke-curley)
  3. moq-rs - moq-rs (draft-14)
  4. moq-rs-draft-16 - Cloudflare moq-rs (draft-16 branch)
  5. moxygen - Meta’s C++ relay
  6. xquic - Alibaba’s XQUIC
  7. imquic - lorenzo-miniero’s C library
  8. libquicr - Cisco’s C++ library
  9. moqtail - TS implementation
  10. quiche-moq - Google’s C++ MoQT (martin-duke, Victor Vasiliev)
  11. moqx - OpenMOQ’s moxygen fork relay (PR #59 merged Apr 11)
  12. mlmtest - Eyevinn moqlivemock interop client (Torbjörn Einarsson, PR #63 merged May 13 17:25 UTC) — supports both draft-14 and draft-16 via DRAFT env var
  13. moqx (client) - OpenMOQ moqx client role (giovanni-marzot, PR #66 merged May 13 17:24 UTC) — adds the client-side image to complement the existing moqx relay
  14. aiomoqt - Python asyncio MoQT client (giovanni-marzot, PR #67 merged May 13 17:23 UTC)
  15. Nokia v17 (via Docker RELAY_URL) - yuyou (Nokia) Docker relay-URL configuration support (PR #65 merged May 13 17:25 UTC) — enables Nokia’s in-house v17 implementation to slot into the matrix

Pending PRs

  • PR #57 (Lorenzo Miniero, open since Mar 19) — Add imquic relay and client docker images.

Current Target

The interop runner currently targets draft-16 for automated testing. PR #68 (Update interop target to draft-18) opened May 18 21:01 UTC by mike-english is staged but not yet merged.

Test Results History

PeriodTotal TestsPassFailSkip
May 24, 2026 00:43 UTC168451220
May 23, 2026 00:42 UTC168461211
May 22, 2026 00:43 UTC168421250
May 21, 2026 00:45 UTC168461211
May 20, 2026 00:46 UTC168381291
May 19, 2026 01:36 UTC168351321
May 18, 2026— (no run)———
May 17, 2026— (no run)———
May 16, 2026— (no run)———
May 15, 2026— (no run)———
May 14, 2026— (no run)———
May 13, 2026105197214
May 12, 2026105207114
May 11, 2026105217014
May 10, 2026105207114
May 9, 2026105207114
May 8, 2026105197214
May 7, 2026105207114
May 6, 2026105207114
May 5, 2026105207114
May 4, 2026105246714
May 3, 2026105246714
May 2, 2026105256614
May 1, 2026105246714
April 30, 2026105236814
April 29, 2026105236814
April 28, 2026105226914
April 27, 2026105236814
April 26, 2026105226914
April 25, 2026105246714
April 24, 2026105236814
April 23, 2026105226914
April 22, 2026105226914
April 21, 2026105207114
April 20, 2026105187314
April 19, 2026105187314
April 17, 2026105187314
April 16, 2026105236814
April 15, 2026105236814
April 14, 2026105217014
April 13, 2026105207114
April 12, 2026105217014
Early April 20269319-2259-6212
Mid-March 202693-9724-25~6012
Early March 20267118-22variesvaries
February 2026 (draft-14)11-149-102-5varies

The jump from 93 to 105 tests (Apr 12) coincides with moqx joining the matrix, adding new client-relay pairs. Transition from draft-14 to draft-16 target occurred in mid-February 2026, which temporarily reduced pass rates as implementations updated.

April 17 regression: 5 tests flipped from pass to fail between the Apr 16 and Apr 17 runs. The regression coincides with the large moqtail draft-16 merges on Apr 14–16 (unified message registry, REQUEST_ERROR consolidation, message parameter fixes) and moq-dev/moq changes landed on Apr 16–17 (broadcast replacement queue, auth module refactor, --cert/--key flag split). Need investigation to identify which pair(s) regressed.

April 18–20: Still at 18/73/14 — no recovery from the Apr 17 regression across three consecutive daily runs. Pair-level investigation remains pending.

April 21: First partial recovery — 20 / 71 / 14 (+2 pass, −2 fail) after four days flat at the regression floor. The lift coincides with ongoing draft-16 fixes in moqtail and moq-dev plus iteration on the SUBSCRIBE_NAMESPACE relay flow in moq-rs PR #157. Still 3 short of the Apr 16 baseline (23/68/14). Need pair-level diff to identify which tests flipped.

April 22: Second consecutive partial recovery — 22 / 69 / 14 (+2 pass, −2 fail vs Apr 21). Two daily improvements in a row; the likely drivers are moqtail PR #175 (subscription timeout 1s → 5s, merged Apr 21) and Suhas’s follow-up forwarding/datagram fixes on moq-rs PR #157 (Apr 21 06:39–08:46 UTC, notably restoring the datagram forwarding rate from 1/sec back to 50/sec). Now just 1 short of the Apr 16 baseline (23/68/14).

April 23: Flat at 22 / 69 / 14 (00:35 UTC run). First day without movement after the two-day Apr 21–22 recovery (18 → 20 → 22). The 1-test gap to the Apr 16 baseline remains. Apr 22–23 implementation work was dominated by spec/refactor PRs in moq-dev (none on main yet besides docs) and moq-wg review activity — no new fixes landed that would move the matrix.

April 24: 23 / 68 / 14 at 00:35 UTC — matrix finally ticks up one pass, matching the Apr 15–16 baseline. A single test flipped fail → pass since the Apr 23 run; pair-level diff not directly exposed in the summary report, but the timing is consistent with the moq-dev/moq hop-based-clustering merge (PR #1322, Apr 23 23:26 UTC) flowing through moq-dev-rs / moq-dev-js docker rebuilds. Still well below the 105-test theoretical ceiling — the draft-17 matrix remains far from complete.

April 25: 24 / 67 / 14 at 00:32 UTC — second consecutive day of improvement and a new April 2026 high-water mark (Apr 15–16 baseline was 23/68/14). One more test flipped fail → pass. This is the first time since draft-17 publication that the matrix has improved on two consecutive days. No moq-dev/moq merges to main happened between the Apr 24 and Apr 25 runs (PR #1322 was the most recent landing); the gain is more likely attributable to ongoing moqtail or moq-rs container rebuilds. Eve of the Apr 27 IETF interim — the matrix enters the meeting at its strongest April reading.

April 26: 22 / 69 / 14 at 00:34 UTC — two-test regression from the Apr 25 high-water mark, breaking the three-day improvement arc (22 → 23 → 24 → 22) and dropping back to the Apr 21–23 plateau. The two flipped tests are not exposed in the summary report. Implementation activity in the Apr 25 02:00 UTC → Apr 26 00:34 UTC window: moqtail merged PR #168 + PR #169 into the draft-16 integration branch (not yet on main, so docker images shouldn’t have changed); moq-dev/moq merged a Python examples PR (#1345) and a dependabot bump (#1347) — neither of which touches the wire path. Most likely a flaky test or an upstream image rebuild for one of the other matrix entries (moq-rs, moq-rs-draft-16, moqx, quiche-moq, libquicr, xquic, imquic). The matrix enters the Apr 27 IETF interim at the Apr 21–23 plateau, not at peak strength.

April 27: 23 / 68 / 14 at 00:34 UTC — one-test recovery from the Apr 26 dip, matching the Apr 24 reading and the Apr 15–16 baseline. Still −1 below the Apr 25 high (24/67/14). Improvement-arc-with-regression: 22 → 23 → 24 → 22 → 23. The flipped test is not exposed in the summary report. Implementation activity in the Apr 26 02:00 UTC → Apr 27 00:34 UTC window includes moq-dev/moq’s two main merges: PR #1340 (wait_for_broadcast / deprecate consume_broadcast — directly affects relay/origin lookups) and PR #1343 (subdomain-based slug routing — affects connection URL handling but not the matrix’s connection URLs). PR #1340 is the more plausible cause for a moq-dev-rs / moq-dev-js image rebuild that flipped one pair. The matrix walks into the Apr 27 interim at parity with Apr 24, neither at peak nor at the Apr 21–23 plateau.

April 29: 23 / 68 / 14 at 00:38 UTC — +1 pass recovery from Apr 28 (22/69/14), back to the Apr 24 / Apr 27 reading. Walking arc since draft-17 publication: 22 → 23 → 24 → 22 → 23 → 22 → 23. Still −1 below the Apr 25 high (24/67/14). The two interim-PR merges (moq-transport #1611, #1609) merged Apr 29 00:03–00:04 UTC are spec-only and don’t trigger an implementation rebuild; the moq-dev/moq merges (#1352, #1353, #1355) merged after the Apr 29 00:38 UTC run, so they couldn’t have driven the +1. Most likely a flaky test or an upstream image rebuild for one of the other matrix entries.

April 30: 23 / 68 / 14 at 00:38 UTC — unchanged from Apr 29. Walking arc: 22 → 23 → 24 → 22 → 23 → 22 → 23 → 23. The four moq-dev/moq merges (PRs #1357 fetch_group + #1350 mTLS HTTPS + #1349 static catalog + #1360 jemalloc) all landed Apr 29 16:08 UTC → Apr 30 00:01 UTC, after the Apr 30 00:38 UTC interop run. Apr 30 spec-only merge PR #1619 (NAMESPACE response name) doesn’t drive interop. Expect possible matrix shift in the May 1 → May 2 window from moq-dev-rs / moq-dev-js image rebuilds.

May 3: 24 / 67 / 14 at 00:38 UTC — first regression after the 4-day +1/day recovery streak (−1 pass / +1 fail vs May 2’s 25/66/14). Walking arc: 22 → 23 → 24 → 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24. Implementation activity in the May 2 00:37 UTC → May 3 00:38 UTC window: moqtail’s PR #180 (separate stream for SUBSCRIBE_NAMESPACE) merged into the draft-16 branch — but moqtail’s docker images point at the draft-16 branch already, so a rebuild could change the matrix; moq-dev/moq’s PR #1366 (flake bump), PR #1368 (doc), and PR #1369 (moqsink EOS) all merged. Luke’s notable PR #1372 (revert of fetch_group + TrackDynamic) merged May 2 21:18 UTC, after the May 3 00:38 UTC report — its effect would only show up in the May 4 run. Most plausible cause: the moqtail draft-16 branch image rebuild flipped one pair from pass to fail.

May 4: 24 / 67 / 14 at 00:38 UTC — flat day (identical to May 3 reading). Walking arc: 22 → 23 → 24 → 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24. The May 2 21:18 UTC PR #1372 revert (Luke removing the partial fetch_group / Subscription API merged Apr 30) presumably reached the moq-dev-rs / moq-dev-js builder before this run, so any restoration-effect from undoing the broken-API change is already baked in — net zero on the matrix. Other implementation activity in the May 3 02:00 UTC → May 4 00:38 UTC window: moqtail PR #145 (umbrella draft-16) picked up 3 commits May 3 (two race-condition fixes + a logging refactor) but stayed on the un-merged branch; no main merges in moq-dev/moq or moq-transport. The matrix has now stabilized at 24 pass for two consecutive days, with one excursion to 25 on May 2.

May 5: 20 / 71 / 14 at 00:37 UTC — major regression: −4 pass / +4 fail vs May 4. Largest single-day regression since the Apr 17 drop, returning to that floor level. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20. Most plausible cause: moqtail PR #145 merged into main May 4 19:23 UTC — the wholesale draft-14→draft-16 migration (216 files, +17,114/−11,744) — image rebuilds for moqtail-relay and moq-dev-rs / moq-dev-js likely flipping multiple pairs to fail. moq-dev/moq PR #1374 (DATAGRAMS Lite05, opened May 4 22:57 UTC) is not yet merged, so does not affect this run. No moq-transport main merges in the May 4 02:00 UTC → May 5 00:37 UTC window. This is the first time in the May arc the matrix has returned to a 20-pass reading — pair-level diff inspection warranted.

May 9: 20 / 71 / 14 at 00:39 UTC — partial recovery: +1 pass / −1 fail vs May 8’s 19/72/14. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19 → 20. Bounce-back to the May 4–7 floor (also the post-PR #145 floor). moqtail PR #193 still open Day +3, so this is not a moqtail-relay rebuild effect; moq-dev/moq main is quiet (no commits since May 7 18:17 UTC, so no moq-dev-rs / moq-dev-js rebuild). Most plausible cause: natural per-run variance / single image rebuild for one of the matrix entries (moq-rs, moq-rs-draft-16, moqx, quiche-moq, libquicr, xquic, imquic) flipping a single test back to pass. Two-day net effect (May 7 20 → May 8 19 → May 9 20) is zero — the May 8 reading was statistical noise, not a regression.

May 14: No new run yet as of this update. The most recent matrix snapshot is still the May 13 00:41:38 UTC reading of 19 / 72 / 14. The ~00:40 UTC daily run for May 14 has not yet published to the GitHub Pages site. Carry-forward: the May 14 reading will be the first matrix snapshot that could reflect a quiche-moq image rebuild post-May 13 google/quiche moqt commits (4 new commits May 13 including vasilvv first moqt-dir commit; combined with May 12’s PUBLISH_OK removal = 6 commits in 48h). If the auto-rebuild propagates and quiche-moq is now ahead of moq-transport-spec (draft-16) used by other matrix entries, expect potentially larger movement than the +1/−1 daily variance.

May 13: 19 / 72 / 14 at 00:41:38 UTC — −1 pass / +1 fail vs May 12 (20/71/14), breaches the post-PR-#145 floor of 20 on the downside. First 19-reading since May 8. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19 → 20 → 20 → 21 → 20 → 19. Implementation activity in the May 12 02:00 UTC → May 13 00:41 UTC window that could drive this: (1) google/quiche moqt landed 2 commits May 12 — “Remove PUBLISH_OK message” (14:23 UTC, commit message: “Part of implementing draft-17/18 PUBLISH in draft-16”) and “Allow fragmented MOQT object payloads” (17:52 UTC); the PUBLISH_OK removal is a wire-format-affecting change and any quiche-moq pair that had been passing on the PUBLISH_OK code point could be expected to flip. (2) moqtail PR #194 merged May 12 20:00 UTC (refactor: remove track forwarding preference dead code, +27/−52) — also pre-cutoff. (3) moq-transport PRs #1625 / #1605 / #1630 merged May 12 20:02–23:07 UTC are spec-only and don’t trigger an implementation rebuild. The matrix at draft-18 publication day: with draft-18 just cut and no implementation tracking draft-18 yet, the matrix is testing draft-16 + draft-14 only; the spec-vs-implementation gap is at its widest. The matrix has been at 20±1 for 9 of 10 May-weekday readings since May 5 — completely insensitive to the parallel spec activity.

May 12: 20 / 71 / 14 at 00:37:28 UTC — −1 pass / +1 fail vs May 11 (21/70/14), back to the post-PR-#145 floor. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19 → 20 → 20 → 21 → 20. The May 11 +1 lift was per-run variance, not a real recovery — the matrix has now been at 20±1 for 8 of 9 May-weekday readings since May 5. No implementation activity in the May 11 02:00 UTC → May 12 00:37 UTC window would account for the flip-back: moqtail PR #193 merged at May 11 22:37:32 UTC (after the run cutoff, so any moqtail-relay builder-rebuild effect from completing the upstream-FETCH series only shows in the May 13 report); moq-transport landed 6 PRs at May 11 21:32–22:02 UTC but these are spec-only (no implementation rebuild path); moq-dev/moq main is Day +2 quiet (no commits since May 9 22:30 UTC). The most plausible flip-back driver is a single-pair regression in one of the May-9-rebuilt moq-dev-rs / moq-dev-js test pairs that was borderline-pass on May 11. Pre-Town-Hall reading: the May 12 MoQ Town Hall opens with the matrix at the floor, having shown no real movement in 8 days.

May 11: 21 / 70 / 14 at 00:42:00 UTC — +1 pass / −1 fail vs May 10 (20/71/14). Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19 → 20 → 20 → 21. Marginal lift above the post-PR-#145 floor, still −4 below the May 2 high (25). Most plausible cause: the May 9 evening moq-dev/moq merges (PR #1338 release-train shipping moq-lite v0.16.0, PR #1392 moq-ffi, PR #1393 cache-eviction 30s→5s) all landed after the May 10 00:40 UTC interop run, so this is the first run in which moq-dev-rs / moq-dev-js docker images were rebuilt against v0.16.0. The May 9 release-train (#1338) and the cache-eviction tuning (#1393) are the most likely flip-drivers: PR #1393 reduces idle memory at the cost of more cache rebuilds, but it also tightens the eviction window in a way that could flip one previously-borderline pair. moqtail PR #193 still open Day +5 (mergeable_state=blocked), so no moqtail-relay rebuild. moq-rs PR #167 still open Day +5, so no moq-rs-draft-16 rebuild. Per-run variance can’t be ruled out, but the timing is consistent with a moq-dev builder rebuild.

May 10: 20 / 71 / 14 at 00:40:03 UTC (gh-pages commit f70964a 01:04:20 UTC) — flat 2nd consecutive day at the post-PR-#145 floor. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19 → 20 → 20. moq-dev/moq’s May 9 evening merges (PR #1338 release-train shipping moq-lite v0.16.0 19:27 UTC; PR #1392 moq-ffi 21:41 UTC; PR #1393 cache eviction 30s → 5s 22:30 UTC) all landed after the May 10 00:40 UTC interop run — any moq-dev-rs / moq-dev-js builder-rebuild effect from 0.16.0 would only show in the May 11 report. moqtail PR #193 still open Day +4 with mergeable_state=blocked, so no moqtail-relay rebuild has touched the matrix. The only other in-window implementation activity is moq-rs PR #167 (Suhas filter-framework, +12163/−2197 unchanged) — open and unmerged, so no moq-rs-draft-16 rebuild either. The May 5 −4 regression remains structurally uncorrected; the matrix has been at 20±1 for 6 of 7 May-weekday readings since May 5.

May 8: 19 / 72 / 14 at 00:38 UTC — −1 pass / +1 fail vs May 7. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20 → 20 → 19. Breaks the 4-day floor downward; new post-NAB low matching the early Apr 17–21 floor reading of 19. moqtail PR #193 (sharmafb upstream FETCH on cache miss, +248/−132, OPEN since late May 6) did not merge May 7, so this is not a moqtail-relay rebuild effect; the most likely cause is another image’s rebuild or natural per-run variance. moq-dev/moq merges in the May 7 06:00 UTC → May 8 06:00 UTC window are PR #1387 (revert-of-revert, type-level only) + PR #1386 (Firefox network stats source) — neither is a wire-format change, so the regression is in implementation pairs rather than a spec change. Reverted on May 9 (returns to 20).

May 7: 20 / 71 / 14 at 00:38 UTC — flat day, third consecutive day at the Apr 17–21 floor reading. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20 → 20. PR #1341 (moq-mux backport) merged May 6 01:20 UTC was expected to potentially affect the May 7 report — no movement observed, suggesting the moq-mux structural refactor preserved wire interop as designed (catalog-side Container::Cmaf { init: Bytes } is the only schema change but is internal to moq-dev’s hang catalog format, not a moq-transport wire change). Other May 6 → May 7 implementation activity in the relevant window: moq-dev/moq merged 3 small PRs (#1382 test-fixture cleanup, #1383 watch broadcast-flap fix, #1385 revert of #1356 insert_track-takes-TrackConsumer) — all post-May 6 00:36 UTC interop run, but #1385 is a same-day pull-back of yesterday’s API change, so the net builder-side delta is zero. moqtail merged 188 (sharmafb upstream-FETCH series, relay code) + #179 (davemevans Firefox docs) — relay code change in 188 (+240 lines) could trigger a moqtail-relay rebuild that would appear in the May 8 report. The post-PR #145 floor at 20 pass continues for a third consecutive day — the matrix is not recovering from the May 5 −4 regression on its own.

May 6: 20 / 71 / 14 at 00:36 UTC — flat day, identical to May 5’s 20/71/14. Walking arc: 22 → 23 → 22 → 23 → 23 → 23 → 24 → 25 → 24 → 24 → 20 → 20. Two consecutive days at the Apr 17–21 floor reading; no recovery from the May 5 regression. Implementation activity in the May 5 02:00 UTC → May 6 00:36 UTC window: moq-dev/moq merged 7 PRs (#1341, #1356, #1377, #1378, #1379, #1380, #1381) — final merge 34dc7c9 at May 6 01:20 UTC was after the May 6 00:36 UTC interop run, so the moq-mux backport (largest of the day) does not affect this report. Earlier merges (PRs #1377–#1380, all merged 17:17–22:15 UTC May 5) could be reflected if moq-dev-rs / moq-dev-js builders rebuilt before the run; none are wire-format changes (API surface tightening, Origin renames, DNS fix, TOML config), so wire interop should be preserved. moqtail merged PR #189 + #191 + #190 (May 5 13:40–14:04 UTC) — moqtail-ts-only changes; Rust relay code unchanged, so the moqtail-relay rows should be unaffected. moq-transport had no main merges. The post-PR #145 moqtail-relay state is the new normal until specific pair-level fixes land. PR #1341 (moq-mux backport) merged after this run; matrix shift could appear in the May 7 report.

Best Performing Pairs

  • moq-rs-draft-16 ←> moq-rs-draft-16: All tests pass (self-test)
  • moq-dev-js ←> moqx: 6/6 pass
  • moq-rs-draft-16 ←> moqx: 5-6/6 pass
  • moq-rs-draft-16 ←> moxygen: 5-6/6 pass
  • moq-rs-draft-16 ←> imquic: 5-6/6 pass
  • moq-rs ←> moq-rs: All tests pass (self-test, draft-14)
  • moq-rs ←> moqx: All tests pass (draft-14)

Getting Started

To add a new implementation to the runner, see the getting started guide.

Related

Graph View

DraftsConceptsPeopleImplDiscInteropOtherTags
DraftsConceptsPeopleImplDiscInteropOtherTags

Backlinks