Skip to content

Vocal Throw M6 P0 — repair wrong-FX Auto-BPM native snapshot and stale-graph readback - #495

Closed
masarray wants to merge 3 commits into
integration/arson-vocal-throw-m1-m5-20261109from
feature/vocal-throw-m6-native-tempo-readback-20261109
Closed

masarray wants to merge 3 commits into
integration/arson-vocal-throw-m1-m5-20261109from
feature/vocal-throw-m6-native-tempo-readback-20261109

Conversation

@masarray

@masarray masarray commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

Vocal Throw M6 P0 — correct native Auto-BPM readback + deterministic beat re-lock

Stacked on #494 integration based on protected current main; preserves M1–M5 and all Harmony, Compressor and Player work. Existing branches untouched; draft until new exact-SHA Windows CTest.

Two reproduced root causes

P0 native snapshot owner: Main.cpp::emitNativeEqSnapshot incorrectly emitted NINE music tempo/beat status fields into the Parametric EQ payload, while emitNativeThrowSnapshot had none. The Vocal Throw editor therefore showed manual/TAP or “NO SIGNAL” even when the engine's MUSIC bus was active/locked. Prior Chromium FAT mocks fabricated a complete Throw snapshot and missed this C++ integration defect.

P0 native CTest red: Integration #494's real Windows run #42613152945 compiled successfully and passed 68/69 tests except NativeMusicTempoSourceOnsetsAndFallback at tests/native_music_tempo_test.cpp:111 recovered.tempo.locked. A weak percussive passage fills the fixed ring with old, unrelated hits. Candidate score only uses onset pairs in the previous 11s, but confidence was divided by ALL stored hits (count_, up to 128) even after those stale hits no longer contribute. Auto-BPM then fails to reacquire when clear backing returns.

Implemented corrections

  1. Move all native MUSIC channel, BPM/confidence, phase-grid, anchor/drop counters to the real arMixBuiltinThrowSnapshot owner; remove unrelated EQ pollution. Gate every readback by current device epoch AND graph revision: obsolete transport cannot show a live lock, stale source resets to 0 and user-visible fallback is Manual/TAP.
  2. Normalize BPM confidence by the count of recent <=11s onset hits, same horizon used by the candidate score; the fixed ring no longer poisons reacquisition. Retain original .28 confidence gate, sparse bounded CONTROL-only candidate scanning, 60–180 BPM range, hysteresis, sample clock and zero-heap ASIO behavior.
  3. Add explicit tests/native_peq_assembly_test.py guard verifying the nine BPM fields are present only in the Throw native C++ snapshot, fenced by epoch/revision, and consumed by the real GUI. Existing native weakBusyGrooveMustReleaseBeatGrid fixture checks dropout/fallback/relock and caught the bug.

Acceptance

  • Fast native and Chromium UI contracts on new SHA.
  • One owner-authorized exact-SHA Windows MSVC/JUCE native CTest on this integration-derived branch, must pass all 69/69 before any green claim. Old 64/64 tests and green fast checks are not full integration proof.
  • Hardware ASIO, microphone bleed, MUSIC bus routing, restart/xrun and live audible singer tests remain separate blockers.
  • No auto merge, no stale data copied to main, no UI-only fake DSP.

Previous handoff #473 (comment)

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: fea39dd
Status: ✅  Deploy successful!
Preview URL: https://c85acc56.arsonmix.pages.dev
Branch Preview URL: https://feature-vocal-throw-m6-nativ.arsonmix.pages.dev

View logs

@masarray
masarray changed the base branch from integration/arson-vocal-throw-m1-m5-20261109 to main November 9, 2026 12:23

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native fea39dd

@masarray
masarray changed the base branch from main to integration/arson-vocal-throw-m1-m5-20261109 November 9, 2026 12:30

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

M6 P0 checkpoint — 2026-11-09 (live branch audit and real CI)

Committed head fea39dd95a3099865b20641f63d57c91d11d4820 (stacked Draft #495 above integrated M1–M5 Draft #494). Older milestone PRs #469/#474/#476/#479 and M5 #493 are untouched. New current-main integration #494 is clean relative to main, but its Windows CTest exact run #42613152945 FAILED 68/69 on test 68 NativeMusicTempoSourceOnsetsAndFallback, tests/native_music_tempo_test.cpp:111 recovered.tempo.locked. This is NOT a green release.

Root cause A — real native GUI readback: nine Auto-BPM/beat-phase fields were erroneously populated by emitNativeEqSnapshot, NOT emitNativeThrowSnapshot in src/Main.cpp. Browser's mocked Throw snapshot hid this. M6 moved the fields to their correct native event and fenced current MUSIC source/BPM/beat anchor with exact device epoch and PCM graph revision. EQ no longer carries foreign source state. Manual/TAP and realtime DSP unchanged. New regression in tests/native_peq_assembly_test.py enforces correct C++ event owner, all nine properties, graph fence and actual JS consumer.

Root cause B — Windows red reacquisition: M5 sparse tempo scorer only counts onset pairs in the most recent 11s. Confidence previously divided by count_ (all 128 remembered hits, many stale weak percussion), so after weak busy groove the strong beat score could never regain lock. M6 normalizes confidence by the matching recent 11s onset count and keeps .28 threshold, existing scoring equation/low-confidence fail-closed behavior, 60–180 BPM, CONTROL-only CPU budget and no ASIO callback allocations. The red weakBusyGrooveMustReleaseBeatGrid native test remains untouched and mandatory.

CI: exact-head fast native #42616334813 PASS; exact-head fast UI #42616334881 PASS. A single owner-authorized exact-SHA Windows #42616675859 is running on this fixed M6 SHA. Do not claim 69/69 until it finishes successfully. Dispatcher-only requirement temporarily retargeted PR #495 to main; base restored to #494 integration branch after dispatch, keeping review diff minimal. No merge or installer change.

Remaining: verify Windows test #68 reacquires; if red, inspect log and fix estimator, not tests. Then reconcile M6 into final integration branch before main, preserve current Harmony/Compressor/Player; perform actual ASIO singer/MUSIC BUS field QA, xrun/CPU, device/WIRE restart, live backing syncopation and PDC tests. This checkpoint is a durable task handoff, not approval for release.

Copy link
Copy Markdown
Owner Author

Repository housekeeping (2026-11-10): Closing as superseded, not as a direct merge. The M6 tempo snapshot/readback and onset-reacquisition corrections were incorporated into the later M1–M11 Vocal Throw source reconciliation (merged #497) and combined M14 (merged #499), including native tempo tests. This older stacked draft is archived, not merged again. Hardware/ASIO certification is still pending.

Code and review history remain on this PR; this closure does not assert a stable/public hardware-qualified release.

@masarray masarray closed this Nov 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant