Skip to content

P0 Coordination Control Tower — One Owner per File, PR Collision Gate and Agent Handoffs #397

Closed as not planned

Description

@masarray

Reconciled 2026-11-08 — historical process superseded. The mandatory GUI/ENGINE/INTEGRATOR/FAT file-owner collision tower is expressly SUPERSEDED by the single product integration owner in current AGENTS.md and merged #418. Keeping its former admission policy marked active misleads new contributors. No PR coordination barrier is restored; branch protection and relevant fast/exact-SHA tests remain. Current issues #412, #192, #15, #389 cover real technical work.

Current main at audit: c57720ea25585c861af89802e1365ffef7d03d17. No statement of physical hardware FAT.


Multi-agent coordination control tower (P0)

One approved /preview/ GUI, one native JUCE PCM engine, one integration owner. This issue coordinates PR work; issue #389 remains the technical GUI/native stitch.

Sole active ownership

  • GUI: site/** and browser presentation; never create another GUI.
  • ENGINE: engine/** and audio DSP/device source, native PCM graph, VST3; sole callback owner PR [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266.
  • INTEGRATOR: src/Main.cpp, CMakeLists.txt, web/approved-native*, native bridge, and shared integration contracts.
  • FAT: hardware evidence, Windows tests, packaging and release qualification, without editing production DSP or GUI.

Mandatory new PR rules

  1. Before coding, inspect current main, open PRs, this issue and AGENTS.md.
  2. Declare HTML comment <!-- arsonmix-workstream: GUI --> (or ENGINE, INTEGRATOR, FAT) and <!-- arsonmix-issue: #NNN --> in the PR body.
  3. Work only on files owned by that lane. For shared source, request a handoff from its owner rather than editing it in multiple PRs.
  4. Existing Windows/Chromium CI will run a cheap PR collision/ownership preflight; new PRs fail on active same-owner overlapping edits. Historical DRAFT PRs through [P0 M2.5] Repository & architecture reconciliation — one owner, no duplicate GUI/bridge/PCM #399 get nonblocking migration warnings; first strict PR is P0 coordination: enforce single-owner PRs and detect parallel edits before heavy CI #400. When a historical PR is marked ready_for_review it becomes STRICT and triggers the existing CI lanes again.
  5. Handoff must include precise SHA, owned paths, producer/consumer contract, CI proof and blockers. One integrator assembles joined Windows application.

Current collision / quarantine

Do not add another CI workflow. Keep GUI on fast Chromium+Pages, engine on native Windows CTest, native seam on both, release on full FAT.
Soft GitHub ownership cannot replace branch-protection required checks; enable those in repository settings when permitted.

Live coordination gate (2026-11-05 WIB)

Activity

  1. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M2.5 actual execution: main merged #396 at 126ab2ef after exact-head Windows+Chromium PASS; closed duplicate implementation #395 and duplicate gate #401; canonical owner guard #400 now contains the unique cross-file engine/bridge domain locks from #401 and strictly rejects a new alternate UI, with PR #400+ gating. This #397 is the ONE control tower; #399 closed as duplicate. Parked #393 GUI archival migration, #270 old engine-heavy FAT, #251 WASAPI overlap; unique tests are preserved in source history. Active owners: #266 PCM/VST3, #398 release admission, #400 coordination, #270 FAT evidence-only after rebase, #225 FactoryFX packaging. Hardware #15 remains PENDING; no claim full LIVE or Windows UAT.

  2. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M2.5 reconciliation checkpoint — joined cross-thread evidence (2026-11-05 WIB)

    Accepted main: 0d23ff904c460f163f2596afe149fafe7f7c8640. Keep this as the baseline for new work; old PR base SHAs are not runtime authority.

    Merged, with concrete evidence

    Current active / blocked merge authority

    Convergence order (one main, not a big-bang merge)

    1. Qualify and merge P0 coordination: enforce single-owner PRs and detect parallel edits before heavy CI #400 policy only when BOTH checks green; no additional CI workflow.
    2. ENGINE owner rebase/extract the real single ChannelId PCM callback and native VST3 leases from [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 onto latest main, with exact Windows native/CTest, no old renderer.
    3. INTEGRATOR alone connects accepted PCM producer to existing arMixConsoleSubscribe/Intent/Ack/Snapshot from approved GUI. Test input→fader→MAIN, actual send/Group/FX Return, VST3, legitimate ACK and same-revision snapshot; reject fake PCM and JS demo updates.
    4. GUI owner separately reconciles P0 — ONE physical GUI source site/preview + same Pages/JUCE assets, no v2 forks #393 physical source normalization with byte-for-byte asset parity.
    5. FAT owner separately reconciles [INTEGRATION/FAT v7 G0] Canonical JUCE WebView ACK/PCM evidence and exact-SHA release gate #270 with an exact-SHA Windows EXE and real audio hardware tests. P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 dry 0FX crackle remains independent stop-ship.

    Hard definition of done: same built EXE uses one approved /preview/, one native JUCE callback, one ChannelId graph, honest VST3/meter/OUT, no duplicate renderer or legacy branch overwrite, combined exact-head native+GUI CI, and physical FAT evidence. No product-ready/UAT claim today.

    This checkpoint is the shared thread handoff. If any owner touches another domain, stop and request a file-specific handoff here rather than create another overlapping PR.

  3. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M3 native-bridge integration collision audit — quantified diff, DO NOT RAW MERGE (2026-11-05)

    Read-only cross-thread reconciliation from current GitHub diffs. At this check, accepted main=0d23ff904c460f163f2596afe149fafe7f7c8640; #396 (approved /preview/ native scope/gestures) and #398 (canonical source/release admission) are merged. The native-only scope __ARMIX_NATIVE_SCOPE__, original finishFaderPointer intent hook, web/approved-native-bridge.js ACK/snapshot handling and full approved PNG buttons are now protected main behavior. Do not overwrite with an old GUI branch.

    Quantified unmerged overlap (GitHub changed-filename lists, not resolved merge conflicts)

    Branch pair Same changed paths Critical examples
    ENGINE #266 × FAT #270 51 src/Main.cpp, src/LiveAudioEngine.h, CMakeLists.txt, engine/core/**, workflow/tests
    ENGINE #266 × GUI #393 5 CMakeLists.txt, ui-smoke.yml, packaging/CI tests
    FAT #270 × GUI #393 5 same native build/CI and tests
    ENGINE #266 × WASAPI #251 4 src/LiveAudioEngine.h, src/Main.cpp, UI diagnostics
    GUI #393 × coordinator #400 2 AGENTS.md, ui-smoke.yml
    ENGINE #266 × coordinator #400 2 both existing CI workflows
    FAT #270 × coordinator #400 2 both existing CI workflows
    FactoryFX #225 × coordinator #400 1 AGENTS.md

    Concrete branch-skew: #393 still contains a 31-line web/approved-native-bridge.js and old scripts/build_approved_native_gui.py without __ARMIX_NATIVE_SCOPE__/native fader commit (current main bridge ~354 lines; source builder ~246). A raw #393 merge would regress the merged M3-A accepted-state GUI. #266 still names web/console-native.html in CMake, rather than the sole approved native /preview/ bundle. #270 still cites retired GUI PR #278 as authority. Individual GREEN CI on these old branches proves no joined compatibility.

    Apply ONE ownership / merge order

    1. P0 coordination: enforce single-owner PRs and detect parallel edits before heavy CI #400 INTEGRATOR policy gate stays the only coordination implementation. No second PR, workflow or parallel source edit. Before merging, confirm latest exact-head Windows + Chromium PASS, fix the review findings for PR-template parked-marker and pull_request: edited, and align docs to strict PR P0 coordination: enforce single-owner PRs and detect parallel edits before heavy CI #400+ (not M2 — approved /preview/ native v6 readback + ACK/snapshot gate (no fake LIVE) #395). Earlier P0 coordination: enforce single-owner PRs and detect parallel edits before heavy CI #400 runs failed domain-collision unit tests; latest branch may differ.
    2. [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 ENGINE: keep ONE LiveAudioEngine callback, PCM/VST3 lease/Show + real ChannelId tests; do not bring web/console-native.*, old CMake, src/Main.cpp, or old release checker into main. Hand off the true native snapshot + device epoch/accepted revisions and v8 intent dispatcher ABI to INTEGRATOR.
    3. INTEGRATOR joins [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 producer with the already-merged M3 FIRST REAL SOUND — wire approved GUI closed state into native bridge; fix fader commit and FX safety #396 /preview/ adapter without introducing a second transport or visual renderer. Gate: exact native input→channel→MAIN PCM; SET_FADER/PAN/MUTE/SOLO and WIRE with ACK→accepted snapshot, meter telemetry and genuine native VST3.
    4. GUI P0 — ONE physical GUI source site/preview + same Pages/JUCE assets, no v2 forks #393 remains PARKED. Extract ONLY genuine site/Pages source consolidation after rebuilding against current main; an owner-authored native bundle builder change must preserve public_from_native(), native-only state scope and approved PNG buttons. Reproduce source hashes/screenshot parity.
    5. FAT [INTEGRATION/FAT v7 G0] Canonical JUCE WebView ACK/PCM evidence and exact-SHA release gate #270 remains PARKED for FAT-only exact-SHA Windows artifact evidence, no engine/Main.cpp/CMake reversion. P0 realtime audio: zero-copy healthy WASAPI capture + bounded discontinuity de-click #251 remains parked and is reconciled only by ENGINE. FactoryFX: integrate ArSonComp as fourth release-watched VST3 with native-safe packaging gates #225 remains separate FactoryFX intake.
    6. No M3 REAL SOUND / product-ready claim until one exact integrated SHA passes both builds, real PCM and physical FAT (P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 crackle still stop-ship).

    This is a conflict inventory, not a request to merge everything. Keep one issue #397 control tower, technical owner #389, PR #400 guard, engine #266, approved GUI main, and FAT #270 only.

  4. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    COORDINATION POLICY IS NOW ACTIVE ON MAIN — 2026-11-05 WIB

    PR #400 MERGED as 8ade8df7e151c421c736ff1c87ed9b0f2e7eef61. This is implemented code, not a handoff proposal:

    NEXT: ENGINE #266 produces real accepted ChannelId/EdgeId+PCM/VST3; INTEGRATOR joins it with exact approved /preview/ GUI; FAT #270 validates real Windows/0FX crackle #15. Keep separate file ownership and exact-sha evidence; avoid micro PRs/duplicate engines.

  5. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    ENGINE #266 → INTEGRATOR selective extraction / M3-B (2026-11-05)

    Actual code PR #402: #402 — latest main 8ade8df7e151c421c736ff1c87ed9b0f2e7eef61 (M2.5+#400 owner guard). The previously-only-on-ENGINE src/ApprovedNativeV8Projection.h is now extracted without realtime code into owned integration/ApprovedNativeV8Projection.h; CMake registers a JUCE CTest in the existing Windows pipeline. No second GUI or graph.

    Critical producer→consumer contract: projectApprovedNativeV8(snapshot, actualNativeBus, installedChannelIdCallbackAdmitted). The explicit third argument must be true only if the installed JUCE callback genuinely rendered PCM with an accepted matching ChannelId graph. Fake JSON, standalone model, or G0 must pass false. Projection never sets productionCallbackLive and never mutates accepted GUI graph on rejected/malformed readback. Long uint64 decimal IDs remain strings. Tested against ENGINE #266 real LiveDryRoutingLab::nativeBusStatus() fields (graphRevision, rosterRevision, channels, edges, controls, inserts, peaks).

    ENGINE #266 ownership unchanged: src/LiveAudioEngine.h, src/LiveDryRoutingLab.h, engine/core/**, genuine VST3 leases/physical callback, epoch validation and nativeBus provenance. INTEGRATOR owns matching src/Main.cpp dispatch and WebView2 only after ENGINE publishes real accepted graph; do not merge obsolete web/console-native.* or branch CMake over current /preview/.

    Status: source PR exists, Windows exact-head build/CTest executing #41908539088. Not yet merged or product-LIVE; #15 physical FAT remains PENDING. Subsequent integrator patch will switch scripts/check_console_puzzle.mjs projection lookup to the owned integration/ path and wait for real callback before wiring admission.

  6. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M2.5 reconciliation checkpoint — 2026-11-05 WIB

    Verified source of truth: main@8ade8df7e151c421c736ff1c87ed9b0f2e7eef61. M2.5 one-owner coordination already landed via #400 (do not recreate coordinator). #395 legacy parallel native M2 was closed; #396 GUI scope adapter is already incorporated in main. Current active INTEGRATOR projection is #402 (readback projection only).

    Live workstream ownership

    Reconciliation order: #403 CI → #402 Windows CTest → one ENGINE handoff for accepted ChannelId controls/PCM → INTEGRATOR joined exact-SHA Windows CI + immutable /preview/ parity → independent #270 FAT. Hardware #15 remains PENDING; no product LIVE or stable release.

    Do not make new parallel M2/native bridge/engine implementations. Exact main/PR SHA must be re-read each session. For all new PRs the gate enforces role+issue metadata and exact/domain collisions; independent Threads cannot see unpushed private work.

  7. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M3-B native control ABI handoff — REAL code, single INTEGRATOR PR #402

    Commit 4163bf3faebf was added to existing PR #402, without starting a second PR/workflow or changing ENGINE source.

    • New integration/ApprovedNativeControlIntent.h: fail-closed conversion of approved /preview/ v8 SET_FADER / SET_PAN / SET_MUTE / SET_SOLO to the actual ENGINE [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 native NATIVE_SET_CHANNEL_MIX / NATIVE_SET_CHANNEL_MUTE_SOLO; GROUP/FX_RETURN fader/pan maps to NATIVE_SET_BUS_MIX only when truly supported.
    • Validates canonical stable decimal-string ChannelId, exact device epoch + graph/roster revision, installed native bus revision, exact target/role, both public supportedOps and native supportedOps, accepted fader/pan/mute/solo readback parity, physical [-60,+10] dB bounds, finite gain; preserves sibling PAN/GAIN or MUTE/SOLO and requestId. Rejects G0, missing/stale callback, unknown IDs, false capability, malformed/unsupported operations without touching graph.
    • Extends existing M3BNativeV8StableIdReadbackAndFailClosedProjection CTest with valid fader/pan/mute/solo conversions, zero-side-effect failure cases and stale/missing native capability. Uses existing Windows runner, no new workflow.
    • IMPORTANT: translator itself never invokes audio, publishes a LIVE flag or fakes an ACK. src/Main.cpp G0 still rejects all universal intents. The actual producer must first deliver admitted PCM and this translator must be wired to the same native dispatcher; exact-head Windows CTest pending.

    ENGINE #266 required next

    Ensure original CH1–10 + dynamic INPUT ChannelIds actually mutate the one installed PCM callback via setNativeInputMix / setNativeInputMuteSolo and produce updated accepted readback, while preserving CH1 vendor/Show/AUX. Only then advertise corresponding public graph.capabilities.supportedOps. The current #266 nativeBus names and /preview/ public names are intentionally different; never advertise SET_FADER/PAN/MUTE/SOLO merely because NATIVE_SET_CHANNEL_MIX exists. Only INTEGRATOR owns Main.cpp/CMake/approved bridge stitching, and #15 hardware FAT remains PENDING.

    This keeps M3 progressing within one owner branch and avoids claiming First Real Sound from C++ fixtures alone.

  8. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M2.5 completion checkpoint — owner-gate hardening merged

    Current main 3c7ea623960067c0a3daac2f81bfc579ac59b81a includes PR #403 (merged). Exact pre-merge PR SHA ed5365ca5a2ecccef46bf783c51dac25acd63ed7 passed existing Chromium CI #41909834153.

    Reused already merged #400 ownership/PR-collision preflight and clarified previously unowned release admission (INTEGRATOR only) versus independent hardware FAT evidence (FAT only). New tests reject cross-owner mutations and distinct-file, same-release-domain collisions. No new workflow, new GUI, engine, or native bridge. Code is present on main, not just a policy issue.

    Outstanding producer/integration work is NOT falsely labeled complete: #402 current 4163bf3f (single integrator projection seam) has Windows exact-head CI pending; #266 066d7b05 is the sole unreconciled, diverged ChannelId PCM producer; #393 GUI source, #251 WASAPI and #270 FAT remain parked; #15 physical crackle FAT remains OPEN. Do not merge raw #266/#270/#393 into main. M3 starts with #266 PCM handoff → #402 integration → joined Windows/Chromium/FAT on exact single SHA.

  9. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M4 parallel execution claim — safe control-plane lane (2026-11-05 WIB)

    BASE: current main@3c7ea623960067c0a3daac2f81bfc579ac59b81a. I am taking INTEGRATOR-only M4 operator routing preflight, without a GUI/PCM fork. Planned owned files: integration/m4-operator-routing.mjs, tests/m4_operator_routing.test.mjs, and narrow scripts/check_console_puzzle.mjs test-registration only. Do NOT change these in parallel until this PR is reconciled.

    Scope of this lane:

    • Pure accepted-snapshot → native-intent planner for quick Route / Route Only / post-fader send using existing APPLY_WIRE_TRANSACTION ABI, exact decimal-string ChannelId/EdgeId, revision/epoch/capabilities, cycle check, duplicate and double-dry protection; all actions remain ACK→accepted native readback, no optimistic GUI/PCM changes.
    • Fail-closed M4 sidechain/PDC policy and acceptance tests: dedicated VST3 auxiliary detector input must be a separate signal kind (NOT audible WIRE); pre-fader/gain and PDC cannot pretend to work before native bus and tap/latency evidence. Define future producer requirements, do not open hidden DSP.
    • Reuse single approved /preview/ and one Windows/Chromium pipeline; zero changes to site/**, engine/**, src/Main.cpp, src/LiveAudioEngine.h, web/approved-native-bridge.js or M3-B reconciliation: extract ChannelId readback projection from ENGINE #266 into approved integration seam #402 integration/ApprovedNative*.* / CMake.

    Observed cross-team blockers: #266 native WIRE decoder currently accepts POST_FADER only; VST3 src/VstBusLayout.h explicitly disables aux bus; graph-aware PDC not present for full live VST3. These are ENGINE-owner follow-up hooks. GUI owner must later map new one-click gestures to the accepted planner, preserving /preview/ pixel parity. In this PR, unsupported action MUST refuse to dispatch, never fake LIVE.

    Definition of this PR: honest, tested M4 integration contract, not completed sidechain DSP or M4 full product certification. Hardware FAT #15 remains PENDING.

  10. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M4 operator contract execution — real PR #404, no overlap with #402/#266

    Repo base: main@3c7ea623960067c0a3daac2f81bfc579ac59b81a. Own INTEGRATOR PR: #404, HEAD ed1165ad8c29de84016de5cbc940c733cf16bbba. Changed only integration/console-v6/m4-operator-routing.mjs, isolated tests/m4_operator_routing.test.mjs, integration/M4_OPERATOR_ROUTING_HANDOFF.md, and existing scripts/check_console_puzzle.mjs test invocation. No src/Main.cpp, CMake, engine/**, GUI site/**, or M3 #402 files touched. CI owner guard passed.

    Executable now: native-accepted snapshot → exact uint64 ChannelId WIRE ROUTE_ONLY / unity SEND_POST candidate transactions; prevents double-dry to MAIN, stale/unknown identities, duplicate cables, graph feedback and non-wet FX Return. Candidate only; existing M2 controller must issue real intent/ACK and compare later accepted PCM snapshot. Absolutely no parallel DSP/GUI or optimistic publication.

    Fail closed: PRE send (decoder currently POST-only), adjustable send gain (readback existence not exact gain), Group/FX Return creation, detector-only VST3 sidechain, plugin latency/PDC are explicitly NOT claimed. ENGINE #266 handoff comment posted with bus layout, detector-only path, timing/PDC and exact test requirements. GUI owner handoff uses existing approved /preview/ appearance and one-/two-click gestures after proper capabilities.

    CI: Initial Windows+Chromium fast source gate failed one test expectation (cycle was masked by input-role error). I fixed validation ordering on current HEAD; on the new HEAD the cheap gate passed in BOTH existing workflows, and exact-head GUI smoke + Windows native build are in progress: Chromium #41912924788, Windows #41912924816. Keep draft until full CI green. No FAT/UAT or sidechain-audio claim.

  11. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M3-C actual INTEGRATOR runtime stitch — commit 427d62466dcb

    On existing #402 only. No duplicate GUI/engine/CI workflow.

    The previous exact-head Windows run 41910167994 compiled the Windows app and 55 CTests; 54 PASS, 1 FAIL at our own newly-added assertion controlIntentTranslationCases line 145 (test accidentally passed a previously projected snapshot instead of a true G0 snapshot). Corrected this test to use canonicalSnapshot(), not a workaround in production code.

    New integrated changes:

    • integration/ApprovedNativeRuntimeSnapshot.h consumes only LiveAudioEngine::status().dryRoutingLab.nativeBus from [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 and first enforces nativeChannelIdPcmActive=true, actual verified direct/open driver, real input/output physical masks, matching positive device epoch, accepted native route, no fault, real sample rate/buffer. G0 / stale / independent-clock / absent nativeBus never projects ChannelId.
    • src/Main.cpp single approved /preview/ snapshot method now projects this installed callback readback on both subscribe and existing once-per-second tick. NO renderer/source change, second callback, fake ChannelIds, or duplicate engine.
    • Projected native PCM graph continues productionCallbackLive=false; its supportedOps=[] until an actually installed native dispatcher is connected. The old G0 arMixConsoleIntent reject remains; this is deliberately honest. Browser will not submit nonfunctional commands as LIVE.
    • Expanded existing Windows CTest with installed-callback provenance and rejection cases.
    • Exact-head Windows run 41913292390 on 427d62466dcb currently executing. Do not merge until PASS and current main reconciliation.

    ENGINE #266 remains bottleneck: current setNativeInputMix/setNativeInputMuteSolo reject original CH1-10 IDs; need ENGINE-owner fix and admitted real VST3/Show/PCM provenance, then INTEGRATOR can install v8 mutating Intent→ACK→later accepted Snapshot. No physical FAT or user UAT today; #15 OPEN.

  12. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M4 accepted-route control-plane evidence — exact HEAD 9c13f59a7f69ce60a2d9fc43c8624bdad456e1f8

  13. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M3 ↔ M4 selective reconciliation gate — 2026-11-05 WIB

    Read-only audit on current main@3c7ea623960067c0a3daac2f81bfc579ac59b81a.

    Recommended sequence / single owner: (1) qualify #404 exact-head Windows; if PASS, merge its narrowly scoped fail-closed contract into main (or rebase/retest first if main moves). (2) INTEGRATOR #402 update from that latest main and run exact-head Windows+Chromium; keep one src/Main.cpp handler. (3) ENGINE #266 supply one installed callback and original+dynamic ChannelId PCM proof, including audible CH1, CH2, 1 dynamic input, fader/mute/pan, wire Route Only/POST send and correct accepted snapshots. (4) On ONE joined exact SHA test Route Only old MAIN edge removed, no duplicate dry path, send gain/tap, changed PCM impulse amplitude, pending/stale/negative ACK rollback and UI approved /preview/ pixel parity. (5) Then M4-B real routing, M4-C optional VST3 detector-only sidechain + graph PDC, only with native lease/bus/latency readback; #15 independent hardware FAT remains STOP-SHIP.

    Ownership: #404 pure routing safety INTEGRATOR, #402 native integration INTEGRATOR, #266 engine PCM ENGINE; FAT prepares independent regression/real audio evidence, not alternative production DSP. Existing CI only, no extra workflow.

  14. 19 remaining items

  15. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    INTEGRATOR #402 — review-driven audio regression fix, no cross-owner edits

    Owned branch: integration/m3b-native-v8-readback-projection-20261105
    Base main: 4a40e7132871ace9fb7ec4b8c0af07aa42630a39 (#404/#405/#407 merged).
    Current head: 039b87462153 (previous qualified c99f60233a1a).
    Review finding: ready-for-review Codex uncovered actual PCM gain-loss risk in integration/ApprovedNativeControlIntent.h: SET_PAN recomputed unchanged gain from display-clamped -60/+12 dB instead of retaining native linear producer gain. Could silence a 0.001-gain channel during a pan gesture. Fixed source to pass through the verified native linear gain on PAN; dB conversion remains only for explicit SET_FADER.
    Regression: tests/approved_native_v8_projection_test.cpp now checks original level 0, 0.001, 4.0 after projection, PAN only preserves exact native gain, explicit fader -60 still silences. Exactly 2 INTEGRATOR-owned paths changed; no site/**, engine/**, native callback, FAT scripts, policy, workflow edits. Inline review discussion acknowledged.
    Evidence: prior c99f6023 Product Qualification #41926205216 SUCCESS (Chromium+Windows JUCE/MSVC/CTest), and promotion's strict owner admission SUCCESS. New 039b8746 exact-head run #41928672962 pending: do not merge before ALL checks PASS and review verified for this latest SHA.
    Other owners: ENGINE #266 remains 59d1ef66 unmerged, no raw historical branch; FAT #406 advanced to 2f7f3b92 with FAT-only paths and no overlap. Native G0 supportedOps=[], intent rejection, physical crackle #15 remain fail-closed.

  16. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-F EXECUTION CHECKPOINT — #402 MERGED, PRODUCER IS NOW SOLE BLOCKER (2026-11-05 WIB)

    Main: f7d2993e4973cf8944cedac570dc87451beed43f (protected, from squash merge #402). Exact PR head 039b8746215349517556616343ad48dd839a8cc8 Actions #41928672962 SUCCESS: coordination, Chromium, native Windows JUCE/CTest, Product Qualification. No major latest-head Codex review findings. Post-merge main Windows run 41938242006 and UI run 41938242509 were IN PROGRESS when verified; do NOT claim post-merge PASS prematurely.

    Real code now on main: integration/ApprovedNativeV8Projection.h, ApprovedNativeRuntimeSnapshot.h, ApprovedNativeControlIntent.h, one src/Main.cpp native snapshot seam, M4 routing safety and original GUI /preview/ unchanged. This is a substantial integrated native readback milestone, NOT live mutating audio.

    Verified hard blocker: main LiveAudioEngine / LiveDryRoutingLab do not yet contain setNativeInputMix / setNativeInputMuteSolo / nativeChannelIdPcmActive / nativeBus (actual methods exist ONLY on #266 historical ENGINE branch). #266 59d1ef66 is 157 ahead and 117 behind accepted main: never raw merge. Existing public supportedOps=[] and rejected arMixConsoleIntent remain mandatory fail-closed behavior until genuine callback/ABI admitted.

    ENGINE #266 owner action: selective source-only current-main PCM callback/CH1–10+CH11/VST3 patch and exact-head Windows CTest; detailed handoff posted directly to #266. INTEGRATOR will then wire translator→verified ENGINE setter→real accepted ACK→later native PCM snapshot, no second callback or browser DSP. No additional GUI work until first audio sound.

    FAT #406 continues independently as exact SHA Windows field kit, physical #15 still STOP-SHIP; previous main FAT artifact exists from older commit but is NOT first real sound validation. Full EXE/ZIP from this exact accepted SHA requires package/Windows run success and separately real physical verification.

    Admin still needed: ruleset #26568033 has only required PR coordination admission; add Product Qualification as second mandatory check, retaining strict up-to-date. No other new governance milestones.

  17. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-G FAT-only ownership and reconciliation checkpoint (2026-11-05 WIB)

    WORKSTREAM: FAT #406, existing PR only. BASE accepted main f7d2993e4973cf8944cedac570dc87451beed43f (native bridge #402 merged); FAT branch 2f7f3b92e63ac907895e2d40d0e2ad30942a8f8b (Product Qualification #41927631811 PASS). This turn is now reconciling FAT-only release/evidence kit to current main, with no new PR, GUI renderer, DSP, native intent or workflow. Exact eight changed paths: docs/FIRST_SOUND_FAT_QUICKSTART_20261101.md, docs/M3E_FAT_RELEASE_ADMISSION_20261105.md, scripts/collect_hardware_fat_windows.ps1, scripts/m3e_fat_evidence_gate.mjs, scripts/package_native_windows.ps1, scripts/verify_native_windows_artifact.ps1, tests/m3e_fat_evidence_gate.test.mjs, tests/native_artifact_packaging_contract_test.mjs. No change overlaps accepted #402's nine integrator paths. A 2-parent non-force expected-SHA commit will preserve existing FAT commits and take all non-FAT paths directly from new protected main.

    ENGINE #266 remains sole PCM owner, 157 ahead / 117 behind current main; cannot raw merge. M3F mutating dispatcher stays blocked until source-only current-main producer and CH1–10+CH11 PCM proof. Ruleset #26568033 admin-controlled: requires only PR coordination admission, request maintainer add existing Product Qualification too. Hardware #15 stays STOP-SHIP. This FAT change is only package/evidence qualification; not a claim of installed PCM, audible Windows EXE or hardware PASS.

  18. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-G FAT-owned execution — exact-main reconciliation + Windows JSON integrity fix (2026-11-05 WIB)

    Workstream: FAT only, PR #406 (existing draft). Base main: f7d2993e4973cf8944cedac570dc87451beed43f containing merged M3C #402, accepted /preview/ and #404/#407. Current HEAD: 24427debaa84a31f478f22713a2ecd0d54cf3547. Eight FAT-owned files, zero overlapping edits against M3C producer/dispatcher, no new PR/workflow.

    Software fix added: optional BOM read for BUILD_PROVENANCE.json + FAT_RUNS.json from Windows PowerShell 5.1 Set-Content -Encoding UTF8, while preserving SHA256 check, exact EXE sourceCommit, malformed JSON denial, explicit hardwareCertified=false. Two new synthetic regressions plus previously verified real powershell.exe AST parser, 44.1/48-kHz real endpoint matching, ZIP member SHA checks. Updated dated FAT runbook: #402 merged readback-only, ENGINE #266 and M3F live producer/ACK still separate, #15 physical crackle OPEN stop-ship.

    Predecessor qualified: #406@2f7f3b92 Actions #41927631811 SUCCESS: coordination, Windows JUCE/MSVC/CTest, Product Qualification. Latest HEAD run: #41940329375, coordination PASS; Windows live tests in progress at posting. UI smoke correctly skipped for independent FAT files. Do not call latest-head fully green before the job completes.

    Final split: mergeable FAT tooling does NOT mean release-ready or native First Sound. Product software BUILD requires exact-sha Windows EXE/ZIP packaged check on a valid main/manual run; supportedOps=[] and no proven original CH1–10+dynamic ChannelId live PCM => native LIVE PENDING; 44.1/48-kHz witnessed 30min 0FX/1FX and hardware #15 => PHYSICAL PENDING. No raw #270 merge or changes to #266, M3F, site/preview.

    Owner follow-through: complete exact-head Product Qualification, then merge FAT source once admission criteria pass; verify actual artifact ZIP in the post-merge main Windows job, retain hardware PENDING.

  19. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-G FAT #406 coordination update — avoid concurrent writes

    The FAT main reconciliation I committed as 835e15557ee9 remains an ancestor of the new concurrent FAT HEAD 24427debaa84. It is not lost. The other FAT operator fixed the Windows PowerShell 5.1 BOM issue and advanced the same PR; this automatically canceled stale run #41939740203 and started exact-head #41940329375. I am relinquishing writes on #406 while that FAT owner actively edits; no force pushes, no duplicate branch/PR, no premature merge.

    Two review findings at PR #406: BOM addressed by new head; PowerShell Summary elapsed-time vs UTC timestamp spoof remains unaddressed as of 24427de. Specific owner-only handoff comment #6646011245 requests strict native PowerShell validation and optional CRACKLE-baseline refusal plus functional Windows tests. Run #41940329375 and review remain PENDING. #15 physical FAT and #266 installed ChannelId callback remain PENDING. Actual main f7d2993 post-merge Windows #41938242006 and UI #41938242509 both SUCCESS, but no First Real Sound claim.

    Admin action still required for protected-main ruleset #26568033: add existing Product Qualification check alongside PR coordination admission; connected GitHub App has no ruleset write action.

  20. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-G FAT #406 × ENGINE #266 coordination, no overlap — 2026-11-05 WIB

    Accepted source: main@f7d2993e4973cf8944cedac570dc87451beed43f; M3C #402 MERGED. ENGINE-only #266 selectively reconciled 28 source/tests at 21141b572521, advanced by ENGINE owner to e15aefec648f7450e6f2d2afa1c6397dd5f68395. This is an ENGINE handoff, not merged or physical PCM evidence; FAT does not duplicate or cherry-pick it. No ENGINE/FAT changed-file overlap.

    FAT #406 exact new head 09421664733f602a367084388439d8f3abbc0df1 (existing ready-for-review PR) contains source-preserving current-main reconciliation, corrected PowerShell ZIP SHA/AST, 44.1/48k 8-mode ledger, truthful physical PENDING, optional BOM ingestion and independent PowerShell 1800s timestamp consistency test.

    Concrete latest CI forensics: previous FAT 5673584b Actions 41941462067 Windows Node 244 PASS / 1 FAIL: isolated field -Action Summary fixture launched PowerShell from temporary test directory and Get-FileHash command was unavailable. FAT-owned fix on 09421664: replace ALL field collector Get-FileHash calls with module-independent streaming .NET SHA256; new test guards file/report hash provenance, existing real PowerShell end-to-end test checks forged timing. New exact-head Actions 41942024097 started; do NOT claim PASS until final.

    ENGINE owner heads-up: old #266@21141b57 Actions 41941454330 failed Windows Node (among others tests/fx_add_contract_test.mjs bus-negotiation count 3 !== 2 and tests/live_graph_contract_test.mjs bypass smoothness callsite pattern); ENGINE has already advanced to e15aefec new run 41941850022. These failures belong to ENGINE, not FAT. Follow exact-head new CI before judging resolved.

    Hard distinction: FAT source/EXE/ZIP checks can PASS separately from installed CH1–10+CH11 real PCM/accepted ACK/snapshot and real witnessed 30-min ASIO/WASAPI 0FX/1FX physical crackle #15, both still PENDING. One approved /preview/, one ENGINE callback. No merge / release before relevant exact-head software checks; no duplicate CI or PR.

  21. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    FAT M3-G #406 latest owner handoff — exact HEAD and independently reviewed failure fixes

    WORKSTREAM: FAT. PR #406 ready-for-review, exact f819570630f827af38490defb3186b23ac1967eb over main@f7d2993e4973cf8944cedac570dc87451beed43f; 8 FAT-owned files only, no producer/GUI/native dispatcher modifications.

    Resolved actual review risks (3 inline threads resolved with code and regression): (1) optional BOM from PowerShell 5.1, (2) -Action Summary requires UTC start/end to match claimed 30-minute time, (3) PowerShell-only mode summary independently validates unique physical capture/render pin ordinals 0..255 and correct direction counts (equal empty pin arrays block). Additional runner-discovered issue Get-FileHash cmdlet unavailable in isolated Windows test: all FIELD collector SHA256 paths changed to .NET streaming SHA256, preserving EXE/collector/report integrity. Real Windows PowerShell synthetic field-journal regression covers these cases, no false hardware LIVE.

    Exact-head qualification: Actions #41942971149 — PR coordination admission=completed/success; native / windows-build=in_progress/pending; Standalone UI and geometry=completed/skipped. All cheap coordination, Node/PowerShell test and /preview/ checks PASS to date; native build/CTest and overall Product Qualification not claimed PASS until complete. Old #406@567358 failed 244/245 Node solely due unavailable Get-FileHash, corrected; older 2f7f PASS not a substitute for latest HEAD. Ordinary FAT PR runs no stage/ZIP verification, so software PACKAGE PASS requires a real post-merge main Windows artifact for the same source SHA. Hardware issue #15 physical FAT remains OPEN/PENDING; native CH1–10+CH11 PCM/ACK is independent.

    Concurrent owner awareness: ENGINE #266 HEAD 74dec9ee79cbac55106f14dee0b3f271683f9069 (draft), INTEGRATOR M3F #408 HEAD 4abb9ba5c6812ba6e23afb4f36cebad5bb06654b (draft). Both separate; ENGINE #266 prior C1083 missing src/ConsoleLiveBridgeV6.h was handed directly to #266/#408, not cross-owned patched. No raw merge #270; single approved /preview/. Keep all lanes exact-SHA and no public release claims.

  22. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-G FAT reconciliation MERGED — exact source admission, package verification in progress

    Merged: FAT PR #406, squash main commit 5f4915680b7a1406c8565620041e517f568a5d7d (parent f7d2993e4973cf8944cedac570dc87451beed43f). Used GitHub exact-head lease f819570630f827af38490defb3186b23ac1967eb. All 8 changed files were FAT-only QA/packaging/docs, no ENGINE, M3F, native JUCE source, CMake, workflow or GUI changes.

    Gate prior to merge: current FAT PR exact-head Product Qualification 41942971149 SUCCESS with Windows MSVC/JUCE/CTest, Node/PowerShell synthetic regressions, ownership and approved /preview/ parity. All three P2 review threads resolved in code: PowerShell 5.1 JSON BOM, anti-forged UTC 30-minute duration, independently verified physical pin ordinals and mode cardinality; replaced optional Get-FileHash dependency with streaming .NET SHA256 in collector.

    Post-merge exact-main hardware-FAT Windows build: 41944776325 for 5f4915680b7a... is currently running. It MUST complete real EXE PE32+, accepted /preview/ resource hashes, actual ZIP contents + SHA and FAT collector/verifier hashes, FactoryFX source locks, and upload experimental hardware FAT kit before this source artifact can be called SOFTWARE PACKAGE PASS. Workflow does not silently publish stable release. Do not claim PASS from mere PR CI.

    Concurrent owners unchanged: ENGINE #266 (one PCM callback) and INTEGRATOR #408 (actual accepted mutating native GUI ACK) stay separately owned and unmerged. First Real Sound native/physical FAT remain PENDING; crackle #15 STOP-SHIP. No physical audio/driver rig was attached to CI, no claims of ASIO/WASAPI clean sound.

  23. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-F actual GitHub code checkpoint (2026-11-05 WIB), product-delivery mode, no duplicate PRs:

  24. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    P0 FIRST REAL SOUND — user-reproduced NO AUDIO, STOP CALLING BUILD A PRODUCT

    Operator Windows screenshot 2026-11-05 08:12 WIB: top native AR MIX Studio, bottom "Stopped | NATIVE G0 - perangkat aktual terbaca; ChannelId PCM/WIRE belum diterima. Edit audio diblokir", right "SIMULATED METERS - NO REAL AUDIO". Operator explicitly reports NO AUDIO. This is a REAL PRODUCT FAILURE, not an optional FAT paperwork item.

    Exact published main 5f4915680b7a1406c8565620041e517f568a5d7d is not an audible product. Current main src/Main.cpp subscribes to arMixConsoleIntent and explicitly calls armix::native_console_m2::reject(args,accepted) for EVERY v6 mutation (verified source around line 203). Engine native setters are not installed on main; packaged Windows workflow #41944776325 PASS only for EXE/ZIP byte integrity, while native GUI still advertises G0 / disabled controls. Do NOT call the ZIP a ready mixer or ask operator to test G0 device settings as the primary fix.

    Active owner PRs (do not fork them):

    Minimum meaningful outcome before any "First Sound DONE": 1 real joined Windows EXE at immutable source SHA, same original approved /preview/ GUI, actual JUCE device START plus nonzero callback PCM and valid source→CH1→MAIN L/R route, real VST3 chain where enabled, fader/PAN/MUTE/SOLO audible and later exact ACK/readback, CH1–10 + dynamic CH11, evidence for 44.1/48 kHz, safe output and no duplicate callback. Separate CI/NATIVE sample proof from real human wired acoustic FAT. Post precise EXE artifact + source SHA + CI + actual video/log/evidence; no simulated meters credited. If real hardware unavailable, PHYSICAL FAT remains PENDING and no stable release.

    Execution priority: solve the three P1 native correctness defects, close exact-head ENGINE Windows gate, qualify stacked #408 with exact producer, then merge in correct order, package one new Windows candidate. No new GUI, no extra QA-only milestone, no new overlapping PR.

  25. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    M3-F FIRST SOUND — Windows testable software package available (2026-11-05 WIB)

    Screenshot field incident: previous Windows package displayed hardcoded Stopped and SIMULATED METERS · NO REAL AUDIO; its NATIVE G0 indicated no accepted real ChannelId PCM. ENGINE #266 and INTEGRATOR #408 were not yet included in that installed build.

    Actual merged product code: main 6084d1e232f829fced708b7c4d8ca3d9522d4b57, ENGINE #266 installed single JUCE ChannelId PCM producer + INTEGRATOR #408 real intent→native setter→ACK/later readback, and approved GUI /preview/. Additional INTEGRATOR code added existing More-menu REAL AUDIO / FIRST SOUND controls: native Audio Setup, select JUCE reported active physical input, explicit CH1 routingSourceSwitch ACK, user-only output ARM/MUTE, physical device footer state (no auto-arm, no JS DSP). Source PR #408 exact-head run 41948677558 passed owner/Chromium/Windows CTest/Product Qualification; merge #408.

    Exact merged source Windows release-candidate artifact: GitHub Actions run #41950281762 SUCCESS, uploaded artifact id 13380763564, ArSonMix-GUI10CH-P012-Verified-Hardware-FAT-Windows-x64. EXE in artifact: ArSonMix-GUI10CH-P012-Windows-x64.exe (9,942,528 bytes). Verified outside CI: outer artifact ZIP passed CRC, BUILD_PROVENANCE.json sourceCommit=6084d1e232f829fced708b7c4d8ca3d9522d4b57; EXE SHA-256 4b0039cc8655caa9e1b26cbd7f8d5a4c71c48584661bef9ca29fc2224690fb83 matches manifest. FAT collector and gate file SHA-256 also match. Post-merge UI 41950281763 SUCCESS.

    Hard distinction: Verified Windows SOFTWARE PACKAGE, not physical audible PASS. Zero-FX hardware crackle #15 OPEN. User must test real ASIO/WASAPI capture/render device, explicit user patch/arm at LOW output volume, then wired 44.1/48k 0FX and 1FX FAT. DO NOT mark stable release or claim measured audio/latency without physical test.

  26. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    Independent GUI ↔ native engine workflow audit (2026-11-06; read-only source review)

    Baseline inspected: main@591444d342492120b5e848d9a0dbea0d33f15efc. This is a source audit, not an installed-EXE SHA audit, hardware FAT, or proof of no crackles. Use one approved generated /preview/ GUI and ONE JUCE realtime PCM authority; do not create another mixer or second callback.

    Source-verified critical gaps on main

    1. Actual runtime: scripts/build_pages.py publishes /preview/ from internal seed and scripts/build_approved_native_gui.py injects web/approved-native-bridge.js; legacy web/mixer-io.js and web/monitor.js are NOT the shipped native control surface. Never implement fixes in unused standalone controllers.
    2. Input / first sound P0: web/approved-native-bridge.js:234-289 still requires hidden Audio Setup → Patch CH1 → Arm output. The visible GUI IN menu is demo in site/preview-v2/index.html:688, while actual per-channel M4 source-switch exists in src/Main.cpp:1457+. Fix only through current P0 Fix cold WIRE admission and native VST3 picker/editor with prepared graph PDC #418; do not duplicate P0 DAW workflow: one native IN selector per strip, no setup wizard, ACK-driven audio #417/P0 Native DAW: 10-channel physical IN, real M4 controls, compact Audio I/O (approved Windows GUI) #416.
    3. WIRE P0: Native GUI sends APPLY_WIRE_TRANSACTION via approved-native-bridge.js:425-465, but integration/ApprovedNativeRuntimeSnapshot.h:138-160 advertises only four mixer ops and integration/ApprovedNativeControlIntent.h:71-72 rejects everything except FADER/PAN/MUTE/SOLO. Prepared engine has LiveDryRoutingLab::applyNativeWireTransaction but it is not admitted through this public dispatcher. No false 'connected' wire, no patching via JS demo.
    4. PFL/cue P0: site/preview-v2/index.html:698 explicitly labels PFL reference-only; web/approved-native-bridge.js:419-423 handles only mute/solo and rejects cue. LiveDryRoutingLab::setMonitorSendGain has independent MONITOR bus groundwork, but this is not a shipped per-channel pre-fader cue flow, and requires verified distinct monitor output pair.
    5. FX picker/insert P0: src/Main.cpp:2253+ has catalog→addVst3 chiefly original legacy CH1; LiveDryRoutingLab::attachNativeInsert excludes migrated M4 lanes and requires a separate admitted route; approved-native-bridge.js:414-424 does not route openPlugin/slotPower/fxOn to a real native insert. GUI picker opening alone is not insert success. Map stable ChannelId/InsertId, authoritative ACK+later FX instance readback and actual callback output.
    6. FX rack header P1: main GUI sectionFx computes slot power using stored individual slotBypass independent of fxOn; parent effective bypass+state restore is proposed in FX rack master power: bypass all effectively and restore each insert's last state (GUI + native engine) #421/P1 DAW mixer: native WIRE/FX rack master bypass + safe optional STREAM output #422, not in main.
    7. Fader/pan ergonomics P1: site/preview-v2/index.html:599-604,1816-1824 uses linear-dB slider travel; source bridge consumeField only sends on change (approved-native-bridge.js:383-407), while native builder injects commit at pointer release (build_approved_native_gui.py:83-89). Verify continuous native fader movement/semantic taper. Native hook intercepts the original GUI pan detent code (site/preview-v2/index.html:1745-1765); center snapping must work visibly AND in PCM.
    8. Unadmitted controls: MAIN/OUT routing, format, phase, SOF/AUX, add/remove track/bus, Show/Scene, plugin actions are variously GUI-demo, native-policy gated, or not mapped through approved control translator. Audit every public active button against accepted native capability, control authority, and callback evidence, not generic UI tests.
    9. Reliability P0: zero-VST crackles are an open physical incident P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15. LiveAudioEngine.h:2007+ has one callback + cadence/over-budget telemetry; instrumentation is not immunity to OS memory pressure, page faults, driver DPC/jitter or independent playback clock starvation. Isolate ASIO primary from optional WASAPI output; record callback-period vs work time, xruns if driver exposes them, ASRC ring high-water/underflows, CPU+RAM pressure, 0FX vs loaded FX.
    10. Capability and release integrity: public GUI, exact Windows EXE, current source, ACK and producer readback, actual speaker/cue/stream receiver must have independent proof. Mark unsupported functions visibly rather than allow demo-state success.

    Reconciliation of overlapping PRs (checked on 2026-11-06)

    Product-acceptance order

    A. Produce one Windows build from an exact SHA; one compact Audio I/O, direct CH01/CH05/CH10 hardware IN, MAIN PCM, verified fader/pan/mute/solo, no hidden Start/Patch/Arm workflow. B. Real WIRE source→MAIN/BUS connect/disconnect/replug + per-edge gain/tap with ACK and callback PCM; never silently change output while live. C. FX picker→correct selected channel→InsertId and VST3 instance, independent slot bypass, parent rack bypass/restore, plugin editor. D. Distinct physical PFL/cue and SOF/AUX + snapshot/recovery. E. ASIO 0FX ≥30 min 44.1/48 kHz, memory-pressure crackle, hotplug, VST3, and optional nonblocking WASAPI stream-to-OBS FAT. Every behavior must be asserted by exact GUI gesture → request → ACK → later readback → real PCM test. Do not claim engine optimized or DAW parity from green UI/CTest alone.

    Scope: No source files, branches, stable, release or deployments modified by this audit. Keep #412/#15/#192 as existing issue owners and coordinate through #397.

  27. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    DAW/open-source comparative architecture review — ArSonMix conformance gate (2026-11-06)

    Scope & provenance: independent study of Adobe Audition, FL Studio, Ableton Live 12, Ardour, Mixxx, and Carla, reconciled with live ArSonMix main@591444d342492120b5e848d9a0dbea0d33f15efc, current #418@2e4a02ce, #421@b3d1570, #422@45d1a98. Research/architecture acceptance document only: no code change, no physical FAT, no Windows production claim. Do NOT create another competing mixer roadmap: docs/PRD_WIRE_FIRST_LIVE_ROUTING_20261104.md is the current operator direction, with docs/ARD_ARSONMIX_UNIVERSAL_REALTIME_CONSOLE_V5_20261101.md, docs/workstreams/ENGINE_PUZZLE_V8.md, and existing #192/#412/#15 owning detailed work. All future implementation must demonstrate code-to-architecture conformance, not merely cite plans.

    A. What mature mixers do and why it matters

    External reference (direct documentation / open source) Reusable architecture principle ArSonMix acceptance
    Adobe Audition routing and monitoring Hardware IN and OUT are per track; monitor I is not record-arm R; independent pre/post sends CH01, CH05, CH10 accept real physical IN on strip; select IN → accepted route → MAIN PCM without hidden Patch/Start/Arm; monitor and record never conflated
    FL Studio mixer and track/PDC Internal/external routing + independent FX slot bypass + parallel send; PDC across routing paths ONE graph owns primary OUT, sends, inserts; delay metadata propagated or admitted low-latency policy rejects/explicitly warns
    Ableton 12 routing, cue, device delay compensation Track I/O and In/Auto/Off monitor are explicit; Cue stereo pair independent of Main; reduced-latency monitoring is a distinct policy with trade-offs No PFL state unless independently routed physical cue PCM; do not force broadcast clock/PDC overhead on singer's ASIO direct monitor
    Ardour signal flow, route source Tracks/buses/processors/sends/monitor use consistent signal-flow and flexible mono/stereo I/O Render unit is runtime ChannelId graph, not special CH1-vs-dynamic FX owner; typed SourcePort distinct from processing layout and FX bus layout
    Mixxx engine guide, EngineMixer One callback, dedicated main/headphone/booth buffers; no blocking I/O/locks; audio source and route independent of GUI One bounded native DSP renderer produces MAIN, CUE, optional stream tap. Parameters use realtime-safe targets; GUI consumes telemetry only
    Carla plugin host, backend source Rack/patchbay are views over independently managed plugin instances and connections Picker→stable ChannelId/InsertId→instance creation/prepare→accepted graph→PCM→editor; never label a catalog selection as an inserted live VST

    Licensing: study behavior/architecture, do not copy GPL source/identical UI assets into ArSonMix without explicit license and compliance review.

    B. Root conformance gaps in actual current main (not a new symptom checklist)

    1. Two conceptual state authorities. The generated /preview/ script owns local tracks/wires/FX state while JUCE has M4/ChannelId snapshots; build_approved_native_gui.py injects event hooks to stop demo mutations. This adapter must become one typed projection of a native SessionSnapshot plus temporary pending gesture state; no second accepted graph in JS.
    2. Two control namespaces, incomplete operation bridge. integration/ApprovedNativeControlIntent.h:71-72 allows only SET_FADER/PAN/MUTE/SOLO, while web/approved-native-bridge.js:425+ emits APPLY_WIRE_TRANSACTION; integration/ApprovedNativeRuntimeSnapshot.h:138-160 advertises only 4 ops. Do not add another per-button handler; define ONE capability-checked command dispatch table with typed operation family (parameter / routing-structural / plugin lifecycle / device lifecycle / show), requestId, exact epoch and revision, correct role validation and independent later readback.
    3. Legacy M4 and dynamic ChannelId special casing. src/LiveDryRoutingLab.h rejects some native insert/source/output operations for original M4 channels (M4_INSERT_USE_EXISTING_RACK, M4_INPUT_NOT_DYNAMIC, M4_INPUT_ROUTE_NOT_DYNAMIC). Single callback exists, but native feature parity and plugin ownership are not unified. Consolidate through existing accepted publisher and lease system, with deterministic M4-to-ChannelId migration/golden PCM parity, not a rewritten second engine.
    4. Structure edits often require ALL outputs disarmed. Valid for current safety but not accepted as the final direct live-operator workflow. Design preflight/stage → resource budget + loop/duplicate-dry validation → revisioned immutable graph → bounded de-click/atomic commit → later PCM confirmation → rollback; never silently hot-repatch speakers, never block callback.
    5. PFL semantics missing at bridge. Current approved cue is GUI reference only; true pre-fader cue must have independent MONITOR/CUE bus with separate verified output pairs; SOLO-in-place and PFL must not share one boolean. Explicitly preserve MAIN audio while toggling PFL.
    6. FX rack lifecycle disconnected. MAIN addCatalog is primarily legacy CH1; nativeHooks.consumeAction has no comprehensive slot/editor chain; no valid live VST3 slot success may be inferred from picker. Accept InsertId and readback. Rack header OFF makes every slot effectively bypassed without erasing individual bypass flags; header ON restores them. Preserve plugin instance/tail/latency and use smooth bypass envelope.
    7. Fader gesture contract incomplete. Source site/preview-v2/index.html has continuous visual pointer movement; native event hook commits at release. Design a realtime parameter lane: deduplicate/coalesce control values at controlled rate, provide immediate local cap while pending and exact accepted value after readback, sample/block smoothing, no graph rebuild/driver restart per fader tick. Use console-calibrated fader taper (unity point + meaningful low-level travel); test audible gain and GUI geometry, including center pan detent.
    8. Zero-FX crackle unresolved. src/LiveAudioEngine.h measures callback cadence and deadline, but success counters are not proof of no physical glitches; RAM percentage is correlation, not causation. Instrument block execution/period (p99/p99.9/max), callback gaps, driver-available xruns, buffer under/overflow, ASRC queue, page faults/DPC (external rig), main/stream independent clock. Preserve primary ASIO callback's no-wait/no-heap rule and opt-in async WASAPI copy.

    C. Canonical signal contract — functional, not cosmetic

    Verified hardware SourcePort → Input Trim/Polarity → PRE-FX tap → FX Insert Chain → POST-FX/PRE-FADER tap → Fader + Pan/Balance + Mute/Solo policy → POST-FADER tap → exclusive Primary OUT + independent Parallel SEND edges → GROUP/FX RETURN → MAIN. A separate PFL/CUE tap takes its configured pre-fader stage to a verified physical headphone pair without changing MAIN. STREAM is an optional post-MAIN PCM copy into a bounded async cross-clock adapter, never an extra native DSP master or latency dependency of ASIO. Every audio port/edge/tap/insert has stable identity and explicit mono/stereo layout; send gain is independent of source fader; routing DAG rejects illegal cycles/duplicated dry paths. Processor reported latency must either be compensated/declared or explicitly prohibited for live monitoring.

    The native SessionSnapshot contains: device IDs/clock/epoch/pins; ChannelIds, roles, order/source/mode; primary edge and parallel EdgeIds with tap/gain; insert IDs/owner/state/bypass/latency; cue/output/stream states; accepted graph/roster revision; PCM-proven meter and health; real supportedOps. UI may represent pending intents, but only acknowledged producer readback changes accepted graph. All commands yield ACCEPTED/REJECTED and later actual PCM evidence where appropriate; no optimistic wire/FX state.

    D. Golden integration gates, required before feature expansion

    Gate 0 (source-of-truth): One generated /preview/ in Pages and Windows; native adapter connects the same DOM to one typed command/snapshot. Exact application executable source SHA documented; no legacy standalone handler considered a shipped fix.

    Gate 1 (first audio): On a fake 4-pin and actual ASIO interface: select CH01 IN1 mono, CH05 IN2 mono, CH10 stereo IN3/4 using only channel strip IN; MAIN 1/2 returns expected distinct test tones. Fader/pan/mute/solo have deterministic PCM effects and an accepted readback; invalid sparse pin, stale revision, device hotplug roll back safely. No operator wizard.

    Gate 2 (routing): WIRE CH01→MAIN/CH05/Group, connect/removal/retarget/route-only and independent pre/post send levels; golden captured PCM matches compiled DAG. PFL CH05 headphone 3/4 changes cue but leaves MAIN waveform unchanged. One graph, no loops/feedback/double-dry. During live operations, staged revisions and continuity proven or a single explicit safety block with reason.

    Gate 3 (FX): On CH01 and CH05 (and dynamic CH11), insert different test plugins, confirm real PCM process count/ownership/editor, reorder/remove, per-insert bypass, parent rack OFF→ON restoring individual prior states. Test mono→stereo bus negotiation, silent bypass continuity, plugin latency admission, independent VST leases; no CH1-special success only.

    Gate 4 (performance and recovery): Zero FX 30-minute ASIO at supported 44.1/48 kHz, vary buffer periods/GUI gestures/hotplug, memory pressure approx 40→70→85% without assuming percentages are causes; compare 2/8 VST and optional secondary WASAPI stream on/off, track period budget and underflows. Native rebuild/CTest/Chromium/packaged SHA + physically audible MAIN, cue and OBS receiver are separate proofs. CI alone cannot establish real-time quality.

    Engineering STOP rules: No additional GUI button, bridge listener, graph feature, or release until its owner has given one end-to-end operation path and golden regression. No new parallel source/controller/PCM graph. If any stage fails, fix the shared command/domain model instead of adding a fallback wizard or special-case. No claim of DAW parity/production/live cue/zero latency based solely on passing build.

    E. Existing PR coordination

    • #418@2e4a02ce is current P0 operator First Sound candidate; exact-head CI remains pending at this review. Review for single-graph conformance and test on actual Windows executable, not just HTML.
    • #421@b3d1570 separately proposes rack parent bypass; do not merge raw into #422 without reconciling the same plugin ownership and accepted-state semantics.
    • #422@45d1a98 stacked P1 draft has Windows MSVC compile failure in ApprovedNativeControlIntent.h (raw undefined) on run 42019107097 and an older #418 base. This is not a releasable WIRE/FX/STREAM milestone.

    This review is intentionally architectural. No patches, extra PRs, main/stable changes, deployment, or physical audio claims were made.

  28. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    Milestone A pre-code reconciliation checkpoint — 2026-11-06 (read-only)

    Current verified baseline: main@591444d342492120b5e848d9a0dbea0d33f15efc. Approved operator GUI is only compiled /preview/, whose internal source is currently assembled by scripts/build_pages.py; Windows JUCE WebView2 uses scripts/build_approved_native_gui.py + web/approved-native-bridge.js. Never patch unshipped standalone web/mixer-io.js as the Windows control authority. AGENTS.md supersedes historical four-owner lanes: one integrated product PR, complete Windows CI, same GUI, one audio callback.

    Actual merge & PR inventory (at reconciliation time)

    Verified architectural collision inside #418 to address before coding Milestone A

    1. Two accepted-state shapes remain: native legacyM4 device/10CH source/control readbacks and admitted graph.channels ChannelId projection. The compiled GUI adapter runs both paintM4Fallback and projectNativeGraphWithoutFocusLoss. M4 may remain an internal implementation, but GUI must consume one normalized authoritative session projection, not independently mutate/choose two models based on scattered conditionals.
    2. Command API fragmentation: arMixConsoleIntent + arMixAction handlers (routingM4Control, routingSourceSwitch, routingChannelPatch, routingFxInsert, various FX bypass) and legacy arMixParameter carry different ACK shapes. Map every operator gesture to one typed request/state transition with explicit device epoch, graph/roster revision, stable ChannelId/InsertId/EdgeId, accepted readback and rollback; retain compatibility internally rather than adding yet another adapter.
    3. FX header state conflict, must not be promoted: #418 web/approved-native-bridge.js uses JS rackSavedBypass and mutates individual slots sequentially to approximate header OFF/ON; intermediate partial ACK/failure and state-of-truth divergence are possible. #421 has an independent native rack-bypass gate that does NOT overwrite per-insert saved bypass. Do NOT ship two policies. Keep FX rack master power: bypass all effectively and restore each insert's last state (GUI + native engine) #421 native design for later integrated FX milestone; no UI-local authoritative bypass preferences.
    4. Adapter growth needs review: web/approved-native-bridge.js grew from 526 lines on main to 1750 lines on Fix cold WIRE admission and native VST3 picker/editor with prepared graph PDC #418; native-build script now injects selector/taper/FX hooks via exact text anchors. Source-anchored build checks are useful, but this must not turn into an unmaintainable second state engine. Reconcile responsibilities and tests rather than further stacking per-feature JS logic.
    5. Source/route transaction correctness: Fix cold WIRE admission and native VST3 picker/editor with prepared graph PDC #418 has live same-format pin switch and control-thread graph repatch for NONE→MONO/STEREO and MAIN/NONE, with control-state preservation. Review physical pin/compact callback ordinal, queued multi-channel edits, 64-bit native identity, previously muted output, and stream-armed structural safety. CI does NOT prove physical audio.

    Milestone A — acceptance scope, not implementation yet

    Goal: ONE real mixer authority, preserving the approved GUI and native DSP architecture; not simultaneous implementation of full PFL, arbitrary WIRE, advanced VST3, and STREAM (those are B/C/D).

    • ONE exact generated /preview/ UI in Pages and packaged Windows; NO alternate runtime, no renderer fork.
    • ONE typed session projection: device/ports/epoch, 10+ actual ChannelIds/sources/processing modes, current scalar controls, primary OUT, supportedOps, real callback meters, and native routing revision; no demo PCM claims.
    • ONE coherent operator command/ACK/later producer readback protocol, with parameter-fast-path separate from structural off-RT transaction; no optimistic browser PCM state, no silent auto-unmute, no stray duplicated legacy-wizard UI.
    • Deterministic end-to-end tests on same Windows package for CH01/CH05/CH10 mono/stereo hardware inputs, independently controlled channels & simultaneous route changes, source→channel→MAIN PCM, fader/pan/mute/solo and MAIN gain/mute, stable identity after channel order swap; deliberate reject on sparse inactive pin, stale epoch/revision, output feedback, driver hotplug.
    • Preserve CH1 external VST3 lease/editor, 4 AUX, Show, device rollback, existing independent monitor/secondary output and realtime no-allocation/no-blocking fences. Native source rebuild must not reset accepted fader/pan/mute/solo or implicitly arm hardware.
    • Exact-head Windows MSVC/JUCE and native CTest, Node protocol+producer fixture, Chromium approved /preview/ interactions, package hashes/provenance PASS; physical ASIO/headphone and 30-min 0FX crackle PENDING until measured.

    Execution decision: reconcile latest main/#419 into existing #418 after its current run completes and HEAD stabilizes; use #418 as only integration vehicle, explicitly separate/defer FX header implementation to the #421 native gate, extract only compatible future work from #422, prohibit a new duplicate PR. If another agent changes #418, re-fetch its head before touching any file. This comment is planning/reconciliation only: no code, branches, merges, CI runs or releases changed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions