Skip to content

P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15

Open

Description

@masarray

Milestone B verified triage — 2026-11-17: P0 / audio-engine / defect / needs-repro. This is classification, not evidence that the latest published build is defective.

Scope / release boundary: Old WASAPI/BT/virtual cable crackling; qualify exact current release before claiming present regression.

Exit criteria:

  • Repeat zero-VST3 44.1/48k WASAPI/ASIO monitoring on exact alpha.26 with devices, driver, clock and 10ms buffer recorded.
  • Capture callback gaps/xrun/packet cadence and audible crackle; isolate clocking from native DSP; pass matched physical A/B before close.

Canonical issue ledger. Preserve field data below.


P0 — Audio crackling even with zero VST3: reliability gate before expanding the engine

Hardware evidence (2026-10-25)

Physical test of experimental main ab732c5 failed: severe crackling at both 44.1/48 kHz, worse at 48 kHz, before loading any VST3. The user's local audio-diagnostics.log (not published here for device/privacy reasons) was fully parsed: 430 one-second AUDIO_HEALTH records over ~7m12s; 298 records had chainCount=0; same Windows Audio shared backend, nominal 480-sample callbacks. An independent virtual capture endpoint and BT-named output endpoint were involved.

Phase Worst cumulative callback time Nominal period Observations
48 kHz, 0 FX 68 µs 10.0 ms 0 slow/over-budget callbacks
44.1 kHz, 0 FX 171 µs 10.884 ms 0 slow/over-budget callbacks
44.1 kHz, 3 FX (ArSonKuPik, FabFilter Pro-C 2, Pro-MB) 2,353 µs 10.884 ms Below measured DSP deadline, still physically unqualified

Longest callback start gap = 21,495 µs, whereas callbackGapCount=0: src/LiveAudioEngine.h counts only gaps greater than 2.3× the nominal period. Callback scheduling jitter itself is not proof of an output xrun, especially for Bluetooth. Instrument endpoint buffer starvation before making claims. 11 audio start/10 stop callbacks include several manual device/rate/channel changes and three plugin loads; addVst3 detaches/re-attaches the callback, which does not explain the pre-FX crackling but is a separate live-operations defect.

Source-level mechanism (confirmed)

  1. ArSonMix src/LiveAudioEngine.h currently has one combined JUCE device callback, measures CPU time and callback gaps but does not expose actual input/output discontinuity counters or clock drift.
  2. In JUCE 8.0.15 juce_WASAPI_windows.cpp, WASAPI shared mode uses the driver-selected default buffer period rather than honoring arbitrary requested callback buffer size. An input reservoir shortage in copyBuffersFromReservoir is zero-padded; the padding path has no dedicated ArSonMix counter.
  3. JUCE WASAPI getXRunCount() in that pinned implementation returns input capture discontinuity flags (AUDCLNT_BUFFERFLAGS_DATA_DISCONTINUITY) rather than complete capture/reservoir underrun plus output render starvation; ArSonMix does not currently publish it.
  4. Two independent capture/render endpoints (virtual input and possible Bluetooth output) can have independent clocks/buffering. Cross-device drift or output scheduler disturbance is a hypothesis requiring controlled isolation, not a proven root cause.
  5. The RT callback repeatedly invokes scratch_.setSize(2, frames, ..., avoidReallocating=true) and creates juce::MidiBuffer locally per chunk. These are worth simplifying/preallocating, but do not explain 0-FX 68 µs callbacks on current evidence.
  6. addVst3, removeAt, moveSlot still detach audio callbacks. P2.5.2 must migrate these to immutable prepared graph handoff. Keep GUI v0.8 unchanged.

Execution order — no P2.6 multi-channel expansion until clean dry passthrough

P0.0: Minimal diagnostic isolation build (do not merge without CI)

  • Add AUDIO TESTS with low-volume native-generated sine or known PCM directly to render output, bypassing capture and VST3: validates output path alone.
  • Test input-only capture statistics independently; retain normal dry-through mode and explicitly distinguish output-only / capture-only / dry-through in diagnostics.
  • Add opt-in local glitch event timestamp + ring stats (no file I/O or UI in callback); output render padding low-water / zero-fill, capture input reservoir underflow, capture discontinuity and driver counts, callback cadence distribution and device lifecycle generation. Do not call timing counters confirmed xruns.
  • Publish getXRunCount() if supported with the correct name captureDiscontinuities; use a backend adapter (JUCE-local extension or WASAPI ETW/IAudioClient instrumentation) where renderer starvation is not exposed. Do not fork JUCE indiscriminately; pin and document patches.
  • Track actual endpoint/native mix sample rate, callback size, endpoint latency vs VST3 algorithmic latency, device mode and independent endpoint IDs. Do not claim the nominal block period equals physical round-trip latency.
  • Add diagnostic marker for a user's crackle report so time windows can be compared without guessing.

P0.1: Hardware A/B root-cause matrix

Use 0 FX first on the same known input signal at matched 44.1 and 48 kHz:

  1. Same-interface wired full-duplex ASIO, where available (single clock).
  2. Windows Audio shared wired output with capture and render on stable hardware.
  3. Virtual-cable capture -> wired output.
  4. Virtual-cable capture -> BT-named output (current failing class of route).
  5. Output-only generated audio -> wired vs BT output; compare capture-only statistics.
    Check Windows endpoint default/communications format, exclusive/sharing mode, enhancements, and BT profile/radio interference as test variables; do not treat any as proven.
    Interpretation: output-only crackle => rendering/BT/backend; output-only clean but dry-through crackles => input clock/capture/FIFO/clock bridge; same-interface works but cross-device fails => independent-clock bridge likely; all paths fail => investigate host/OS driver.

P1: Long-term, modular 48-kHz-first design (after P0 measured)

DeviceHost (ASIO / WASAPI)
   ├─ CaptureEndpoint (device clock A) -> bounded timestamped FIFO
   │     -> drift estimator -> optional asynchronous SRC / rate matching
   └─ RenderEndpoint (master clock B) -> preallocated AudioEngine::process()
                -> ChannelDSP[12] -> MasterBus -> Output/Stream
Control Thread -> Preparer Worker -> immutable PreparedGraph
                -> atomic callback-boundary publication -> deferred reclamation
Audio Thread -> atomic snapshots / bounded telemetry -> diagnostic worker -> WebView2
  • One full-duplex hardware interface = direct path, no unnecessary resampler; different devices = clock-domain bridge only where needed.
  • Master playback clock drives DSP; maintain adaptive FIFO fill, bounded rate correction (in parts per million) and controlled underrun concealment. Define explicit failure mode if devices cannot remain synchronized; never introduce global blocking locks.
  • Callback noexcept, fixed-capacity/preallocated buffers, no dynamic allocations, mutexes, I/O, plugin creation, WebView2, or blocking waits. No blind changes to native rate or buffer while a graph is pending.
  • RAII lifecycle with device generation token; prepare/resume/suspend/stop, single-owner plugin life cycle, safe graph retirement. Plugin-specific calls may violate RT rules: measure, isolate or degrade gracefully.
  • Separate DeviceHost, ClockDomainBridge, RealtimeMixerCore, PluginGraph, PluginLifecycleWorker, AudioTelemetry, NativeUiAdapter into testable modules, rather than growing one LiveAudioEngine.h.
  • Preserve GUI v0.8 web/app.js, web/styles.css, and offline standalone exactly; status adapters only.

Validation/acceptance gates

  • CI: deterministic fake input and output clocks with ±100/±300 ppm drift, 44.1↔48 kHz, 5–30 ms scheduling stalls, mono/stereo mapping, buffer wrap, zero-fill concealment, device change mid-graph, and 12,000+ graph handoffs; tests must assert bounded memory and no callback thread ownership violation.
  • Windows CI must PASS MSVC Release, native CTest, UI Chromium and preset/VST3 contracts.
  • Physical FAT: no audible crackling on supported wired dry path with 0 FX at 44.1 and 48 kHz over ≥30 min, monitor capture discontinuity and true render underrun signals (or honestly mark unsupported). Repeat with plugin chains and device changes under well-defined muted/lifecycle state.
  • Document Bluetooth output as high-latency/non-live-monitoring mode unless hardware tests qualify a device.
  • Keep main@ab732c5 as reproducible experimental checkpoint. No hardware-ready/public-live claims until physical FAT.

References

Priority: P0. Dry, zero-FX audio is unqualified. This supersedes starting P2.5.2 FX edits or P2.6 channels ahead of reliable I/O.

Activity

  1. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    Product-owner clarification — Smart and reliable across devices (mandatory design principle)

    Normal users must not have to engineer buffer sizes, sample clocks, Windows driver quirks, or sample-rate interoperability. The A/B hardware matrix is internal diagnostic/FAT only, not the normal product workflow.

    Required adaptive product behavior

    • Auto-discover endpoint capabilities and select a stable capture/render profile first, not an aggressively low-latency one.
    • Prefer one shared-clock ASIO/full-duplex device for low-latency monitoring; select direct signal path, avoiding unnecessary sample-rate conversion.
    • Support independent-clock routes (virtual cable capture + wired/USB/BT render) through explicit clock-domain architecture, bounded elastic FIFO, drift estimator and asynchronous SRC when needed. Measure sustained FIFO level and effective rate correction, prevent systematic underflow/overflow. The exact algorithms and thresholds must be evidence-backed and benchmarked, not magic constants.
    • Negotiate rates/formats (including 44.1/48 kHz), map mono/stereo correctly, preallocate all device/graph buffers.
    • Adapt output buffering and recovery to endpoint behavior without changing buffer parameters from inside real-time callback or creating jittery oscillating profile changes.
    • Treat Bluetooth as potentially high-latency and variable, not intrinsically low-latency. Clean playback is the first target; communicate unavoidable monitoring delay transparently.
    • Device disconnect/hotplug/change must use a tested state machine, generation tokens and a safe muted recovery path, no out-of-callback data races or access to retired VST3 nodes.
    • Device-health reporting distinguishes measured render starvation, capture discontinuity, clock drift and DSP deadline misses. Zero counters without complete instrumentation must never be advertised as proof of no glitches.
    • GUI v0.8 stays visually locked. Setup should default to AUTO / STABLE, with Advanced/manual overrides only for troubleshooting and professional users.

    Acceptance standard

    1. Supported combinations should render no audible crackling with zero FX under realistic conditions on sustained hardware FAT at 44.1 and 48 kHz.
    2. Verified graceful fallback/degradation, documented capability boundaries and clear notices if a particular broken/incompatible driver cannot provide continuous sound. Never promise universal physical guarantees, zero latency, or guaranteed BT performance.
    3. CI includes simulated multi-device clock drift, jitter and format mismatches plus real wired/virtual/BT hardware matrix. Test automated profile selection, underflow recovery and avoidance of buffer thrash.
    4. Finish dry path acceptance before further P2.5.2/VST3 feature expansion. Preserve approved GUI v0.8.

    This is a core engine requirement, not an optional quality-of-life feature.

  2. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    Architecture matured — 2026-10-25 (PR #17)

    The device-agnostic strategy is now specified as one canonical ARD + one implementation blueprint, replacing vague "optimized" claims and historical v0.3 wording:

    Five hard choices: measured AUTO/STABLE profile rather than blind buffer tuning; direct shared-clock fast path with no extra SRC; separate clocks use bounded FIFO / drift estimator / streaming ASRC only after backend timing feasibility; VST3 affinity-safe prepared-graph lifecycle; provenance-labeled discontinuity vs timing telemetry.

    Execution gates: P0-00 output-only/capture-only/dry-only measurement; P0-01 independent callback/clock feasibility; P0-02 ASRC+fake clocks ±100/±300 ppm; P0-03 deterministic policy+rollback; P0-04 actual hardware FAT; only then P2.5.2/P2.6. CI success is not hardware audio acceptance. GUI v0.8 remains visually locked.

    The PR is documentation and CI enforcement only. No engine performance repair is claimed before P0 implementation and measured FAT.

  3. added a commit that references this issue on Oct 25, 2026
  4. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    Architecture v1.0 accepted into main — 2026-10-25

    Merged PR #17: #17
    Merge SHA: 74b421ff29e4be92b923aca98722d816712eba89
    Recovery checkpoint: checkpoint/smart-audio-ard-v1-20261025

    Canonical documentation now in main:

    Documentation+CI contract only, no code implementing adaptive clock bridge yet. Preserve separate P0 hardware FAT and no-live-certification status. PR head passed Windows MSVC/CTest and Chromium. Post-merge CI triggered on main; check status before marking main checks green.

    Next engineering slice: P0-00 diagnostic output-only + input-only support detection + DRY 0-FX repeatable telemetry, not P2.6 features.

  5. added a commit that references this issue on Oct 25, 2026
  6. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    P0-00 diagnostic implementation merged — 2026-10-25

    PR #19: #19
    Merge SHA: 29b9b0fb01b6e9977f242341e7e3f0c839e9b84a
    Checkpoint: checkpoint/p0-00-native-isolation-20261025
    Runbook: https://github.com/masarray/arsonmix/blob/main/docs/P0_00_AUDIO_ISOLATION_RUNBOOK.md

    • Real output-only route closes capture endpoint; constant-space 440Hz sine at -30dBFS, 8ms smooth ramp, no per-sample trig or allocation.
    • Real capture-only route closes output endpoint and fail-closes/silences if the backend cannot isolate as requested.
    • Dry-through bypasses VST3 execution; output-only and capture-only are diagnostic test modes only.
    • Temporary setup is restored, not persisted; cross-mode reconfigure mutes output; restoration errors retain original setup for retry.
    • MARK CRACKLE and typed provenance report actual reported backend xrun field when available versus estimated callback timing, with render starvation explicitly unsupported in current generic JUCE adapter.
    • Native Windows PR CI GREEN: MSVC Release + CTest 3/3, with Release-active amplitude/finite/ramp checks; Chromium GUI GREEN. All native GUI v0.8 visuals retained. Post-merge CI triggered from main.

    NOT FIXED / GATE STILL OPEN: actual hardware output crackling; verified render starvation; independent-device drift and adaptive ASRC; AUTO/STABLE policy; P2.5.2 live-safe FX graph edits. Next P0-01 is backend clock capture/render feasibility and hardware A/B using P0-00 annotated logs. Do not close this issue or mark live-ready.

  7. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    P0-00 post-merge verification GREEN (main 29b9b0f) — 2026-10-25

  8. added a commit that references this issue on Oct 25, 2026
  9. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    Consolidated P0 Smart Audio engineering infrastructure merged — 2026-10-25

    Merged: PR #21, main@c3191148d203b944df3ad8d199d0b2fc79399293
    Checkpoint: checkpoint/p0-smart-audio-diagnostic-clock-core-20261025
    Handoff: P0 Smart Audio diagnostic & clock bridge documentation

    Implemented and validated in PR Windows Release MSVC + 6/6 native CTest + Chromium GUI:

    1. SIGNAL CHECK → COPY FULL REPORT: native Windows clipboard, device API and reported formats/latency, host callback p95/p99 budget distribution, driver xrun provenance, plugin timing, live route, latest 48 events + 24 important records incl. MARK CRACKLE. No disk reading or DSP-thread I/O.
    2. RealtimeHealthHistogram: bounded RT atomic buckets; handles >100% overflow without inventing a 100% upper ceiling.
    3. AdaptiveClockBridge prepared experimental core: lock-free SPSC FIFO, 48-tap 1024-phase FIR, bounded PI clock drift compensation, ±100/300 ppm simulated independent clocks and 44.1↔48 kHz test. The core is not yet in the live JUCE joined device callback.
    4. AudioProfilePolicy deterministic planner that recognizes only positively verified shared-clock hardware, ranks supported sample rates/buffers and requires an independent-stream adapter for cross-clock paths.
    5. AGENTS + README update: no manual hardware tests for every tiny engineering change; consolidate P0 then one acceptance campaign.

    P0 issue stays OPEN. Actual crackling is NOT fixed/confirmed resolved. Next phase is real independent WASAPI capture/render adapter feasibility, backend discontinuity telemetry, staged clock-bridge/live AUTO policy integration, performance/fidelity and device lifecycle validation. A clear single hardware FAT comes after these integrated milestones, not as a blocking step after each PR. Post-merge Windows/Chromium CI initiated for commit c3191148.

  10. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    Post-merge artifact and CI acceptance — exact main SHA c3191148d203b944df3ad8d199d0b2fc79399293

    • Windows Release CI: https://github.com/masarray/arsonmix/actions/runs/40188984892 — PASS; native CTest 6/6 PASS: CapabilityAudioProfilePolicy, AdaptiveDualClockBridge, RealtimeHealthHistogram, P0DiagnosticTone, RealtimeControlPrimitives, NativePresetZipRoundTrip.
    • Chromium GUI: https://github.com/masarray/arsonmix/actions/runs/40188984889 — PASS.
    • Main Windows artifact: https://github.com/masarray/arsonmix/actions/runs/40188984892/artifacts/12597223676 — ZIP integrity verified, EXE 8,395,264 bytes, SHA-256 ff38b0304fe64ba6a4a061e7b9beb4b988fe7124c91da4ccbcf187c38dfaa8bd.
    • Recovery checkpoint: checkpoint/p0-smart-audio-diagnostic-clock-core-20261025 -> c3191148.
    • Diagnostic workflow: open SIGNAL CHECK -> COPY FULL REPORT -> paste into engineering chat. For user hardware acceptance, defer until core + independent native clock adapter + AUTO/STABLE lifecycle are integrated. Real crackling is still unqualified, even with green simulated clock CI. Keep issue open.
    • Production integration outstanding: JUCE independent capture/render backend, reported output-starvation telemetry, SRC quality/alias benchmarks and calibrated buffering, real device auto selection/fallback, hardware FAT. Do not describe the current unused adaptive core as fixing Bluetooth or virtual-cable crackling.
  11. added 2 commits that reference this issue on Oct 25, 2026
  12. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    2026-10-26 — Live independent Windows clock adapter merged to main

    Merged PR #35 #35
    SHA 8a1daf5b83d02df18e3ab4dd869f5d006db5441c
    Recovery checkpoint/p0-live-dual-clock-adapter-20261026
    Implementation https://github.com/masarray/arsonmix/blob/main/docs/P0_INDEPENDENT_JUCE_CLOCK_ADAPTER_2026-10-26.md

    Now implemented: actual separately opened Windows Audio capture + render JUCE managers/callbacks, render-master bounded SPSC FIFO + prepared polyphase FIR clock compensation, negotiated-period-aware FIFO headroom/rejection, selected microphone/output channel-mask preservation, direct-mode rollback on device error/rate shift, native SMART CLOCK opt-in, backpressure/drift counters in COPY FULL REPORT. Direct ASIO remains without SRC. Three automated inline P1/P2 findings resolved. Latest PR head Windows MSVC Release + CTest 10/10, Chromium GUI PASS. Added 480-frame capture / 1440-frame render continuity regression.

    Outstanding before P0 certification: device-policy activation and recovery hysteresis beyond opt-in, actual output render-starvation signal beyond generic JUCE, cross-clock fidelity/latency benchmarks, real 44.1/48 kHz capture/BT/virtual/ASIO hardware FAT. Under no circumstances close the P0 issue or claim all devices are crackle-free from GitHub CI. Main post-merge Windows/Chromium runs were triggered; check exact SHA when done. Product owner requests consolidated end-of-milestone hardware test and one-click report, not repeated micro hardware trials.

  13. masarray commented on Oct 25, 2026

    @masarray
    OwnerAuthor

    P0 recovery hardening merged — 2026-10-26

    PR #39: #39 (merged)
    Main merge: f9d4059621f49f5eb1a34daa864e42ac5f787092
    Recovery checkpoint: checkpoint/p0-direct-recovery-safe-20261026
    Durable handoff: https://github.com/masarray/arsonmix/blob/main/docs/P0_DIRECT_RECOVERY_HANDOFF_2026-10-26.md

    • After independent device loss, safely restore the prior DIRECT endpoint using control-thread-only schedule: initial attempt and up to 4 retries at 250 ms / 1 s / 3 s / 8 s. Generation-bound, no RT file/OS operations, no endless reconnect flapping or automatic risky return to SMART CLOCK.
    • Startup/failover now verifies that actual prior capture and render endpoint names match, active I/O channels exist, and manually chosen channel masks are preserved. An empty JUCE API error alone is not accepted as a successful restore.
    • A healthy independent SMART CLOCK session does not mistakenly invoke recovery just because the saved DIRECT rollback token is present.
    • If the original endpoint is permanently absent, explicit AUDIO SETUP safely abandons its rollback token, keeps output muted, reattaches host callback for the user-selected replacement, and prevents persistence of an empty transitional setup.
    • COPY FULL REPORT exports retry state, attempts, remaining delay, generation, last error and pending token. Release note corrected from obsolete offline-only ASRC claim.
    • Release-active unit tests + source contracts added. PR-head Windows MSVC Release CTest 11/11 PASS and Chromium PASS; both inline findings resolved. Post-merge CI initiated on the exact main SHA.

    P0 #15 remains OPEN. Required before full P0 qualification: verified stable device identities / AUTO-STABLE orchestration, render-backend starvation evidence, long-duration SRC fidelity/latency/perf and real 44.1/48k wired/virtual/Bluetooth acceptance. No user hardware test per PR; consolidate final FAT with SIGNAL CHECK → COPY FULL REPORT. No live-ready or universal crackle-free claim.

  14. 48 remaining items

  15. masarray commented on Nov 1, 2026

    @masarray
    OwnerAuthor

    FAT v6 integration gate — no acoustic qualification claim

    Draft PR #270 now adds offline validation of native PCM/ACK evidence and exact source/CI/packaged EXE provenance, plus negative test cases (double dry, cross-input, monitor leakage, master mute, stale ACK). These are synthetic validator tests, not a field capture or a fix for 0FX crackle. Existing good baseline Windows artifact remains a 10CH/4AUX FAT candidate. Hardware #15 stays OPEN.

    Required hardware run before clean-sound signoff: low-level wired OUTPUT_ONLY_TONE versus CAPTURE_ONLY versus actual 0FX DRY_THROUGH ≥1800 seconds at logged backend, actual negotiated sample rate/buffer and exact capture/render pins; audio marked events, COPY FULL REPORT, capture flags/zero-head candidate deltas, raw/post-FX/MAIN PCM and physical MASTER mute. Repeat true same-interface ASIO against VB-Cable→Maono Windows Audio to investigate independent clock/resampling without inferring DAC XRUN from capture flags. On integration candidate separately verify independent IN1→CH7 and IN2→CH2, real VST3, Group, parallel Return, independent Monitor, hotplug explicit re-ARM and Show. Do not close #15 or call sound clean from CI/trace alone.

  16. masarray commented on Nov 1, 2026

    @masarray
    OwnerAuthor

    Physical FAT intake instrument now packaged on dedicated CI candidate lane

    Integration PR #270 4a54c9a adds ArSonMix-Hardware-FAT.ps1 into the Windows x64 distribution and validates its SHA against BUILD_PROVENANCE.json and the ZIP. On Windows (once exact-HEAD CI successfully uploads), run Verify, then choose actual native SIGNAL CHECK mode and run Start/Stop for OUTPUT_ONLY_TONE (no input), CAPTURE_ONLY (no render) and DRY_THROUGH_0FX (both endpoints; >=1800 sec) with low wired headphone volume. The field journal records START/STOP timestamps, observed backend/rate/buffer/actual callback pins, reported crackles/CLEAR, full diagnostic report SHA, and witness. Summary writes JSON with hardwareCertified=false and unresolvedIssue=15, regardless of human result; it cannot control output ARM or inspect speaker sound. Physical comparison still needed for same-interface MAONO ASIO vs independent VB-Cable/WASAPI output; capture flags do not directly prove DAC XRUN. Candidate is fixed 10CH/4AUX and must not be represented as universal v6. No sound-clean claim.

  17. masarray commented on Nov 1, 2026

    @masarray
    OwnerAuthor

    2026-11-01 physical screenshots: First Sound route UX is a separate blocker from #15 crackle

    Screenshots supplied by the hardware operator show Windows Audio 48000Hz/480 samples, capture CABLE Output (VB-Audio Virtual Cable), playback Realtek HD Audio 2nd output, one enabled input pin and CH01 apparently STEREO with both L/R selecting IN1, even while NATIVE ROUTE ACK and OUTPUT ARMED labels are displayed. The truthful signal status is NO RAW CAPTURE PCM and NO MAIN PCM. This is not a hardware crackle measurement: the capture source is currently silent, and the stereo UI pin selection is not physically valid. Native monitor.js preflight already rejects duplicate stereo pins; display must not label that map LIVE.

    UX correction PR #275 makes mono single-pin First Sound explicit, one channel patch per action, advanced matrix hidden, and false LIVE labels rejected by browser UI. Exact Chromium CI current #41270290564 pending; this is not yet a packaged EXE.

    FAT next step: ensure a playback application explicitly emits to Windows playback CABLE Input, then capture endpoint CABLE Output should show RAW PCM on the verified driver pin. Realtek is the renderer, not the VB-Cable feeder. Distinguish no-input-PCM from the prior reported crackle on nonzero capture PCM. After audible input/MAIN is established, resume low-volume 0FX 30min crackle isolation, output-only tone, and native report before #15 closure.

  18. masarray commented on Nov 1, 2026

    @masarray
    OwnerAuthor

    First Sound M4 physical candidate — verified GUI, Windows build gate still pending

    Screenshot-driven GUI owner PR #275 source 60a1fb9bef062d4cb9dc5d71ae326aa1aac3fd76 has exact-head Chromium/Playwright PASS 41270440729. It fixes UX: one input row, mono if only one enabled capture pin, distinct stereo L/R, MAIN L/R selectors, Advanced collapsed/no crosspoint DOM until opened, clearer capture/ACK/no PCM guidance. Integration PR #276 HEAD 70c8a485b87638b6012866edd7971d6966a3c781 uses the exact tested GUI snapshot plus native 10CH/4AUX FAT package, and its own Chromium/Playwright 41270572187 PASS. Windows 41270572198 is still building native C++/CTest/package (do not claim artifact available until zip verifier/upload). Native MixerRoutingPlan already rejects duplicateStereoInput, reinforcing that screenshot L=R must not be represented LIVE. Never infer physical sound or close #15 from green browser/native CI; operator must verify RAW PCM→MAIN and 0FX wired audio independently. Universal #266/#271 drafts stay separate; no new DSP/GUI owner.

  19. masarray commented on Nov 3, 2026

    @masarray
    OwnerAuthor

    Integrated native M5 GUI+ENGINE physical FAT candidate — evidence/provenance only

    Joined draft PR #321 combines latest production GUI main 1e1b76a87a9fa8581aaeae4b16a7c0e37d38967e with ENGINE #266 9d0ecabbdd3ba2b1629a8c3917ee64e0b6ae139d, preserving original GUI/CSS/WIRE and the single native M4 10CH+4AUX callback. Exact joined SHA f1855d8d839d665bdfda6b53d002d37ecec8ad5f Windows MSVC/CTest/packaging 41561801600 PASS and Chromium 41561801788 PASS. That original merged CI could not publish ZIP because the upload gate allowed only ENGINE branch.

    Latest PR #321 commit 745dcbbfa333624e0e4c8ae7000b2d335cbab50f adds branch-limited ZIP upload and carries workstream #270's source- and ZIP-SHA256-witnessed ArSonMix-Hardware-FAT.ps1 offline field collector (Verify/Start/Stop/Summary, OUTPUT_ONLY_TONE, CAPTURE_ONLY, wired DRY_THROUGH_0FX ≥1800 seconds; local report hash, exact backend/rate/buffer/input/output pins, human audible observation/witness). The collector never arms output or changes drivers and always writes hardwareCertified=false. Exact HEAD Windows 41562795114 native CTest/PE/ZIP/upload is still running at comment creation; UI 41562795128 PASS. Do not distribute the latest ZIP until all native steps PASS; DO NOT close #15 or assert 0FX crackle is fixed. Actual 44.1/48 kHz wired ASIO/WASAPI side-by-side with physical PCM, mute, full report and witness remains release blocker. Bluetooth/cross-clock paths are not approved low-latency monitoring until demonstrated.

  20. masarray commented on Nov 3, 2026

    @masarray
    OwnerAuthor

    M5.3 joined physical FAT evidence — source committed, exact SHA pending CI

    PR #310 newest joined SHA: c687b4859e8a22853712df7fbfc9893311dea622, integrating without modifying any GUI #271 UI/CSS/WIRE ownership. Uses verified ENGINE #266 native ChannelId realtime SOURCE+MUTE/SOLO and zero-latency VST3 gate. Physical PCM producer remains canonical M4 until full acceptance.

    New real operator FAT deliverable: The existing offline PowerShell field collector from PR #321 has been selectively grafted into the newer combined GUI+ENGINE PR #310, without copying PR #321's older GUI, runtime or incompatible verification scripts:

    • Packages ArSonMix-Hardware-FAT.ps1 in same Windows ZIP with the verified EXE and read-only HARDWARE_FAT_GUIDE.md.
    • Source SHA-256 of collector must match packaged script, BUILD_PROVENANCE.json.fatCollectorSha256, and extracted ZIP entry digest. ZIP verifier fails closed on missing/tampered collector; existing GUI/FactoryFX source hashes and original Windows workflow preserved.
    • Operator -Action Verify/Start/Stop/Summary records actual native readback backend/rate/buffer/pins, endpoint clock class, 0FX wired ≥1800s, OUTPUT_ONLY_TONE and CAPTURE_ONLY, reported crackle, witness, COPY FULL REPORT SHA. It cannot ARM hardware or certify sound; hardwareCertified=false, unresolvedIssue=15 always.
    • Offline packaging contract extended to assert script, manifest+ZIP SHA guards, mode list, and human-only certification status. Windows exact-SHA run 41579304652 and Chromium run 41579304658 pending completion; no artifact claim until verifier/upload pass.

    Architecture reference: console Group primary OUT, parallel PRE/POST FX Return and independent Monitor require graph-aligned latency, retained vendor VST3 identity and control-owned Show recall (Waves eMotion LV1 / Ableton Live / Avid Pro Tools). The new ChannelId FX lease admission rejects 64-sample nonzero VST3 latency until true PDC exists. No universal production PCM cutover, physical audio PASS or stable release claimed. NEXT: unified M4→ChannelId callback authority with CH1 external/Orbit + Show/editor ownership and real GUI ACK + native meter; physical #15 stays P0.

  21. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    M3-E physical crackle FAT kit prepared (not a hardware result)

    Independent FAT PR #406 now provides SHA256-bound Windows field recorder + read-only evidence gate. It requires separately witnessed 44100 and 48000 Hz, OUTPUT_ONLY_TONE, CAPTURE_ONLY, 30-minute wired 0FX, 30-minute matched 1FX, actual device/buffer/physical pin identity, and full REPORT SHA256 from SIGNAL CHECK. Both device-specific 0FX and 1FX must match; all records bind to identical Windows EXE and source SHA. No direct listening or hardware capture was conducted in CI; this is an evidence kit, NOT resolution of crackle.

    IMPORTANT provenance: WASAPI getXRunCount() can reflect capture discontinuity flags; callback gap and DSP deadline do not prove DAC render underrun. Real independent wired hardware/reconnect/Show/VST3 evidence is still required. #15 remains OPEN STOP-SHIP. Do not use synthetic regression results to label clean audio.

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

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions