Skip to content

P0 LMS native engine — crossover sections → matrix → safe multichannel ASIO output #636

Open

Description

@masarray

Milestone B verified triage — 2026-11-17: P0 / lms / feature / planned. Classification ≠ completion of product feature or physical FAT.

Scope / release boundary: Safety gate before enabling any physical multichannel loudspeaker output.

Exit criteria:

  • Prove fail-closed matrix/crossover/port identity: no unassigned pin can emit, no silently retargeted ASIO outputs, safe device restart and muted uncalibrated outputs.
  • Qualify mono/stereo crossover frequency/phase, routing and bounded output DSP with native PCM + physical multichannel field tests before exposing live output.

Canonical issue ledger. Preserve historical evidence below; no speculative merging or rewriting approved audio.


Objective

Implement real native ArSonMix Loudspeaker Management (LMS) beneath the approved /preview/ GUI shipped by UI draft PR #635, WITHOUT introducing a second mixer, second PCM callback, new renderer, or competing device manager. Harmonice is only the separate internal vocal harmony plugin.

Authoritative signal path

existing ChannelId BUS / MAIN → LMS system source → stereo/mono N-way CROSSOVER (sections) → section-to-output MATRIX → per-destination OUTPUT DSP → validated ASIO hardware pins

  • One crossover calculates each section once, with fan-out to multiple destinations; no duplicate processing.
  • Logical SectionId, SystemId, OutputId, verified PortId and device epoch are stable typed identities. Never use physical port ordinal or track index as durable storage identity.
  • Matrix routes section→port explicitly. One section source per output initially (multiple sources to one output only through an explicit mixer with measured headroom, not hidden crosspoint summing).
  • A design may have more logical outputs than currently available physical pins; unmatched mappings persist as Unassigned/Offline, never silently bound or auto-armed.

Native RT architecture (not a JS audio engine)

  • Prepare immutable crossover graphs and coefficient banks on non-audio threads. Realtime JUCE callback uses preallocated channel buffers, precomputed filter state and bounded delay rings.
  • Support at least 1–4 way stereo/mono and scale by actual processing budget for larger N; validate crossover frequencies against current sample rate/Nyquist; HPF/LPF filter-type/order contract (e.g. Butterworth/Linkwitz-Riley, avoid mislabeling LR as Butterworth).
  • Per-output parametric EQ, polarity, gain, configurable delay, calibrated limiter with gain-reduction telemetry. Never promise zero latency or speaker protection from an uncalibrated dBFS threshold.
  • Structural routes, filter family, way-count, device epoch and buffer changes transactionally preflight, ramp/quiet the affected physical outputs, publish atomically with ACK + accepted snapshot. Rejection leaves previous safe graph intact.
  • No heap allocation, filesystem, lock, plugin creation, JSON, UI callback, or blocking inside realtime audio processing.
  • No automatically bypassed speaker safety. Reset/reconfigure must NOT briefly send full-range to HF outputs.

Visual UX and session

  • Reuse only canonical assembled /preview/ (not web/index.html alternate); five tabs System / Crossover / Matrix / Processing / Alignment.
  • Crossover ways created first, Matrix patches way outputs second. Output DSP inspector changes exactly one destination or deliberately selected link/group.
  • Speaker drag/alignment is a geometric draft; require an explicit native Apply to change audio. Show reference distance, time, acoustic-measurement caveats, and accepted delay readback.
  • Integrate system+sections+patches+safe processor states into existing native .smix session ownership and device-independent restore. Hardware identity mismatch leaves routes unbound, no silent reassignment.
  • UI send typed intent → bridge → control-thread validation → native publisher → PCM graph → ACK → same-revision snapshot. No optimistic LIVE badges.

Tests / real acceptance

  1. ASIO with real 2/4/6/8 hardware outputs: 3-way stereo sections feed exactly the designated ports; each output measured, others silent.
  2. Dynamic 2→3→4→2-way changes: no unexpected full-range on tweeter, stale routes or audible pops; repeated device/buffer/sample-rate switching retains logical session safely.
  3. One LOW section fans to two physical sub outputs with independent delay/limiter; no doubling unless explicitly requested.
  4. GEOMETRY preview never affects audio until accepted explicit Apply; sound-speed conversion has deterministic tolerance, no inferred phase alignment.
  5. Rejected output pins, invalid frequencies, insufficient delay memory, CPU budget or unsafe limiter state fail closed and preserve old accepted graph.
  6. Native CTest for crossover frequency response/phase, delay ring, limiter behaviour, graph publication, no RT allocation; cross-layer UI ACK/readback test; one FAT build when user-test executable is required.
  7. Preserve current mixer WIRE, FX/VST3, session/open/close, ASIO monitoring, and only one callback. Do not public-release on green CI alone without actual speaker/driver FAT.

Reference workflow: Allen & Heath AHM Speaker Processing creates ways and requires patching each way to sockets; Powersoft ArmoníaPlus separates system design, configuration and tuning.

Dependency: PR #635 establishes safe offline UI/logic but intentionally DOES NOT implement native acoustic DSP, PCM routing, hardware protection, or .smix persistence. This issue is the vertical native milestone, not a request to ship an unqualified UI-only feature as working hardware.

Activity

  1. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    2026-11-13 LMS M2 engine checkpoint (implementation, not physical activation):

    • Delivered on draft PR feat(dlms): 5-workspace speaker DSP, safe .smix drafts and explicit ASIO remap (M12) #635, exact HEAD 7a0170c0ae509e31f45069028dc967477b8b83cf, reconciled against current main 9c17e926cb6a8edc79941fb82fde91b3c7328adc.
    • Added portable C++ engine/lms/PreparedSpeakerSystem.h with real LR4 24dB/oct crossover section PCM, exclusive verified physical port matrix mapping with source fan-out, preprepared per-output 8-band peaking EQ, fractional delay, gain/mute and DIGITAL sample ceiling.
    • Session epoch, graph revision, current callback channel count, exact verified native output identities, speaker-profile acknowledgement, Nyquist, delay and memory budget are prevalidated. Rejected candidates never replace an accepted prepared DSP graph. No runtime heap allocations in process(); synthetic PCM fixture exercises eight outputs / four stereo ways / three EQ bands each for 1500×256f.
    • CMake defines two actual CTests; fast native Linux gate compiles/runs the same production sources, PASSED; fast UI /preview/ contracts PASSED. One exact-SHA Chromium + one exact-SHA Windows native milestone dispatched, both in progress at this comment; no duplicate FAT.
    • NO claim of physical speaker protection: current hard digital peak ceiling is not a calibrated lookahead limiter and cannot substitute for voltage/power/excursion protection. NO claim of ASIO output yet; kernel is not wired to accepted native callback.

    P0 next required vertical slice: bind logical source to existing accepted MAIN/BUS ChannelId and canonical PCM publisher; prepare exact native verified hardware pin map (no DEMO), reserve output ownership, route section PCM in the SOLE JUCE callback, expose typed LMS control+ACK+same-revision snapshot to the five-page /preview/ GUI; calibrate true limiter/protection, .smix persistence, and real 2/4/6/8-output ASIO FAT before allowing active speaker output. Protect current WIRE/VST3/main output and cross-thread FX work.

  2. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    2026-11-13 M3 audited product checkpoint for LMS, delivered on existing draft PR #635 (not a parallel PR), exact head 268c2a945f4b37475e15acdda45c4bcbdb83fc47 reconciled with main=54702c2796b6713dc6a936276b9c5c11b0f97135:

    Actual code delivered

    • 5-page task-focused /preview/ GUI now uses calculated digital LR4 crossover section frequency curves (not invented rectangles), section→output exclusive matrix and output Processing EQ/Level/Protection tabs with interactive 8-band parametric magnitude visualization. Alignment displays positions and distances, computed geometric delay, and requires explicit confirmation to COPY those offsets only into disconnected DSP DRAFTS.
    • Native CONTROL-only engine/lms/NativeSpeakerOutputReservation.h: device epoch, published ChannelId graph revision and MAIN source identity, real output mask, exact callback ordinal identity and sample rate; reject MAIN/MONITOR/EXTERNAL pin collisions, wrong device, stale revision, duplicate and inactive output. No physical output authorization from digital hard clipping or browser limiter switch.
    • C++ tests/lms_native_output_reservation_test.cpp plus CMake/fast Linux CTest gate, extended JS model/interaction regression, new actual two-viewport Chromium tests/pages_lms_five_workspaces_smoke.py with screenshots. Previous V7 hardware-key transient DOM race now waits for stable attached element.
    • Early fast UI/native checks PASS and the dedicated exact-source LMS Chromium full flow already PASS in run 43398639697. Full historical GUI qualification and Windows MSVC/CTest for exact SHA are being checked; do not mark whole PR green prematurely.

    Remaining P0 work: implement and qualify a true native speaker limiter/protection profile, wire PreparedSpeakerSystem into the SINGLE accepted JUCE ChannelId PCM snapshot, reserve/replace existing physical MAIN output ownership without simultaneous fullrange, publish typed UI intent→ACK→accepted snapshot and .smix, and prove actual multi-output ASIO hardware FAT. This PR remains DRAFT and no speaker hardware is armed.

  3. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    2026-11-13 LMS M4 implemented on draft PR #635, exact reconciled HEAD b8124c737c3e32404f5934e8a8f54c27b60d2d86, main f76ce45bfb8b1df6d8068cc39224b6bf052a6a3a.

    • Added optional CONST read-only post-FX MAIN PCM observer to the existing DynamicNativeSessionPublisher::render; it never re-runs the mixer/VST and does not change the accepted physical output pin map or audio buffer contents. It is pre-master hardware gain and has no GUI authority yet.
    • Added native C++ end-to-end tests/lms_native_postfx_main_witness_test.cpp: accepted ChannelId VOICE→MAIN with hosted FX called ONCE per block, feeds actual output PCM into the preprepared LR4 stereo 4-way kernel using isolated diagnostic memory, 8 band outputs, 250 blocks, zero callback heap, stale epoch does not call observer, physical buffers untouched. This is an in-memory witness, NOT ASIO FAT.
    • Native OutputPatch.polarityInverted and ParametricBand.enabled now match the Processing page's polarity and bypass controls; new real PCM assertions test inverted sign / bypass / gain response.
    • Corrected a REAL PFL key rendering regression in approved GUI: HTML uses data-hw-kind="pfl", while hardware SVG renderer and LED CSS had incorrectly retained cue. All three owners (HTML / renderer / CSS) now align. V7 Chromium test asserts the engraved PFL SVG instead of the former invalid selector. The click handler action remains cue.
    • PASS in fast Linux native at -O2 -Wall -Wextra -Werror: LR4/PEQ/Delay, 8-output 1500×256f zero-heap stress, exclusive ASIO port guard, and REAL accepted ChannelId post-FX PCM witness (logs run 43403016335). On latest merged main SHA the fast native & UI gates also PASS. Full Chromium and Windows MSVC/CTest exact-SHA qualification dispatched once each and running at https://github.com/masarray/arsonmix/actions/runs/43403568282 and https://github.com/masarray/arsonmix/actions/runs/43403647960 .

    P0 STILL OPEN: a trusted native operator LMS intent→ACK and same revision snapshot, fixed identity binding in .smix, the full single JUCE callback post-master/PHY integration, calibrated peak/RMS/thermal loudspeaker limiting and real 2/4/6/8-output ASIO field FAT. No output authorization or PR merge.

  4. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    2026-11-13 Native LMS M5 delivered on the existing draft PR #635, exact SHA 6e0cc76f46399d2d6dbccf2a38a79b04b6df67e1 (main/base f76ce45bfb8b1df6d8068cc39224b6bf052a6a3a).

    M5 concrete code: DynamicNativeSessionPublisher::acceptSpeakerDraft accepts a typed speaker plan only after validating the CURRENT accepted native ChannelId MAIN, graph/device epoch, sample rate/quantum and native verified output pin inventory; all physical outputs must be disarmed and all speaker patches remain muted. It prepares actual DSP off-RT, then emits typed native ACK and full CONTROL-only readback including graphRevision, designRevision, physical pin ordinals and hardwareApplied=false. It retains the last accepted draft on rejection (stale CAS revision, wrong source, port collision, invalid DSP, unprotected unmute), and invalidates on structural PCM graph commits, device unpublish and physical external pin repatch. It adds no second callback, no active amplifier writes and no fake LIVE state.

    M5 release-mode test tests/lms_native_draft_ack_readback_test.cpp is compiled/executed under Linux -O2 -DNDEBUG -Wall -Wextra -Werror and added to Windows CTest. Exact native fast CI run 43409922575 PASS with M2/M3/M4/M5 PCM gates; fast UI run 43409922571 PASS. V7 PFL/SOLO actual SVG regression test now reads <text> with text_content() rather than HTMLElement-only innerText.

    Still P0: connect a typed WebView→native runtime request producer + accepted snapshot subscriber; ensure .smix persistence stores only offline safe configs and must never auto-arm speakers; prove actual callback output ownership + calibrated voltage/RMS/peak loudspeaker protection and real ASIO multichannel FAT. Until then PR #635 remains DRAFT and output disarmed. Full Windows and full Chromium exact SHA qualification pending.

  5. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M6 implementation contract (engineering reference, not yet implementation)

    Research baseline: Allen & Heath AHM documents that crossover processing output/Way is distinct from full-range outputs and must be explicitly patched to the correct physical socket, with an output limiter per Way. Dolby SLS Audio processor guidance warns that speaker protection requires manufacturer-approved driver limit and amplifier output voltage calibration, not a dBFS-only toggle. References: https://support.allen-heath.com/hc/en-gb/articles/40185032073873-Configure , https://support.allen-heath.com/hc/en-gb/articles/40185199750161-Channels , https://professional.dolby.com/siteassets/products/sls-audio-products/sys-300/sys300a_processor_settings.pdf

    Next implementation acceptance gate (M5 now provides a safe immutable native control readback but NO physical sound):

    1. Typed WebView LMS Design Intent -> native validation -> same active device/graph revision ACK -> saved/readback visual state; do not emit APPLIED unless a single JUCE audio callback has actually committed the physical graph. A cached or imported .smix must always open speaker outputs muted/disconnected; no automatic arm on restart or device switch.
    2. Exactly ONE hardware output writer: the existing accepted ChannelId MAIN/BUS post-processing PCM enters the prepared N-way crossover once per source, then PEQ/polarity/delay/gain and per-driver protection. Full-range MAIN shall NEVER be written to the same loudspeaker pin as a filtered way. Independently occupied MONITOR/BUS/direct output pins must reject admission; native output mask and epoch are read from actual device, not browser labels. Every physical output has a unique owner, stable OutputId+PortId and revisioned readback.
    3. Per-output certified-by-measurement limit profile must include driver/manufacturer recommended peak/RMS voltage (or acoustic power limit with known impedance and assumptions explicitly displayed), actual measured amplifier gain/sensitivity, output calibration to DAC full-scale RMS, HPF bounds, and required peak/RMS attack/release. Validate amplitude scaling at actual hardware DAC; a sample hard clamp or a UI flag is not a thermal/peak/excursion protection model. Never infer amplifier gain/speaker impedance from the GUI.
    4. Future DSP: RT-bounded peak envelope and time-averaged RMS/thermal limiter as appropriate; fixed-size preallocated lookahead only if the latency budget permits. Explicitly report latency; do not claim true-peak protection without true-peak measurement/oversampling. Validate power-on, silence, feedback, subsonic, clipped input, output retarget, device hot swap, unplug/replug and stale revision fail-closed. Null/unverified profile and uncalibrated hardware always remain muted.
    5. FAT: use a dummy load and instrumented measurements before connecting expensive speakers; verify one 2-way mono, 2-way stereo (4 pins), 3-way stereo (6 pins), 4-way stereo (8 pins), independent L/R, no full-range duplicate, correct muted defaults, correct driver voltage limits, and native meter/ACK rollback. Do not label synthetic PCM CTest as hardware FAT.

    M5 code and test on draft PR #635 remain deliberately DIAGNOSTIC/CONTROL only. Current fast Linux native and fast UI checks PASS at 6e0cc76f46399d2d6dbccf2a38a79b04b6df67e1. One exact-SHA Full Chromium and one native Windows qualification are running; do not merge/arm hardware based on fast checks.

  6. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    2026-11-13 LMS M6 delivered in existing DRAFT PR #635, exact reconciled HEAD 141aa19af739b178b1f02fea213aee6fa1388c5c, base 27ffb25802287b81285e61e068ff4e87ae1f6101 (Guitar FX layered-artist CMake preserved).

    Implemented production control vertical slice: Real JUCE JuceActivePortSnapshot + accepted ChannelId MAIN/EXTERNAL pin ownership produce one trusted ASIO output inventory. NativeSpeakerDeviceBinder discards WebView fake callback ordinals, requires exact device/graph revision, explicitly disarmed outputs and no MAIN/MONITOR/EXTERNAL collisions. LiveAudioEngine::validateLmsDesignOffline checks DIRECT/ASIO/diagnostic/reconfig gates before and after native revisioned DynamicNativeSessionPublisher::acceptSpeakerDraft; device drift revokes the saved draft. A typed bounded NativeLmsWebIntent parser, Main.cpp WebView listeners, and existing compact five-workspace GUI support Inspect ASIO → explicitly select verified pins (clears old patches) → Validate muted design → pending ACK → independent native readback. The GUI only shows VALIDATED DRAFT for matching deviceEpoch, graphRevision, designRevision and section route, never APPLIED. Public /preview/ has no fake native bridge and remains DESIGN ONLY.

    Tests: six actual portable native PCM/port tests PASS including binder and M5 ACK; Node simulated WebView 3 scenarios PASS (browser fallback, real inspection/readback, reject/armed device). Fast UI and Fast Native on exact SHA PASS (runs 43416772098 and 43416772117); dedicated Windows JUCE decoder+readback CTest LMSNativeJuceIntentSchemaAndDisconnectedReadback and full executable compile running in one exact-SHA native milestone 43416938783 at comment time. CSS V25/V26 historic ledger preserved with exactly one audited PFL LED selector normalization; regressions retained.

    P0 STILL OPEN: physical output PCM publication inside the ONE JUCE graph (post-MASTER), calibrated driver/amp peak/RMS/thermal limiter, safe startup/disconnect + .smix persistence, verified actual 2/4/6/8-pin ASIO dummy-load FAT. No speaker outputs armed; PR remains draft, not ready to merge as hardware product.

  7. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M7 engineering audit/fix (2026-11-13), PR #635 head 9d22970c46be0c3706e5f0a9dd7c8de6a6233380:

    Root cause: M6 browser post-ACK sameNativeReadback() previously compared only way count, cutoffs, physical route identities, section and mute. The JUCE inspection() native projection omitted gain/delay/polarity, limiter intent and PEQ parameters entirely. This permitted a genuinely different output DSP configuration to be labelled VALIDATED DRAFT despite native ACK/readback mismatch — particularly serious for loudspeaker settings.

    Code now implemented: src/NativeLmsWebIntent.h serializes bounded per-output gainDb, delayMs, polarityInverted, limiterEnabled and all per-band frequencyHz/q/gainDb/enabled into the native inspection snapshot. site/preview-v2/assets/lms-workspace.js only marks VALIDATED after full strict parameter parity and disarmed device proof, in addition to the existing epoch, graph, design revision, route and crossover checks. It never labels hardware as APPLIED. Includes controlled mutations against gain, delay, polarity, limiter, peq insertion/removal, missing fields and unauthorized unmute in tests/lms_native_bridge.test.cjs; JUCE native source roundtrip assertions in tests/lms_juce_web_intent_test.cpp.

    Verified: exact HEAD fast native source/UI contract PASS run 43417975374, fast UI PASS run 43417975365, targeted Windows preset workflow PASS run 43417975407 (not the entire LMS Windows CTest). One exact-head Full Chromium and one dedicated Windows native qualification requested /arsonmix-ci ui|native 9d22970c46be0c3706e5f0a9dd7c8de6a6233380; pending. No hardware ASIO speaker authorization, no claim of calibrated protection. PR stays draft.

  8. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    LMS M8 — native post-master diagnostic PCM integration (2026-11-13)

    Product PR: #635, head 6be711496168ebef8c04f02299ab8c43ecfb2d8c (base 27ffb25802287b81285e61e068ff4e87ae1f6101). Source commits: ce7ee2176dde1b3c00c79d419534b4c29fccdc17 + compile/memory-budget correction 6be711496168ebef8c04f02299ab8c43ecfb2d8c. This remains a DRAFT / NO PHYSICAL SPEAKER OUTPUT.

    What now actually runs

    • Existing sole LiveDryRoutingLab::process PCM callback reports the exact accepted ChannelId render revision, then feeds already-applied master gain, ramp, MAIN-only mute and mono policy PCM to DynamicNativeSessionPublisher::processSpeakerDiagnostic. Prior M4 observer was pre-master.
    • On CONTROL, acceptSpeakerDraft verifies actual physical ASIO reservation and prepares a second diagnostic-only instance of the real M2 LR4/matrix/output-DSP kernel. All unmuted diagnostic destinations are buffers privately preallocated in that instance; the accepted hardware design remains fully muted and unarmed, without any ASIO pointer passed to diagnostic processing.
    • One gate-based RT lease protects diagnostic DSP lifetime; old state is revoked on accepted graph/physical patch/device change. The sink requires matching device epoch + actual rendered PCM revision. Diagnostics report per-pin PCM peaks and post-master L/R peaks; telemetry explicitly marks hardwareApplied=false.
    • Tests in tests/lms_native_postfx_main_witness_test.cpp cover actual one-pass ChannelId FX source, post-master gain scaling to separated LR4 outputs, L/R isolation, 220-block zero-heap processing, no physical buffer writes, stale epoch/revision refusal and revoke. No new renderer, mixer or callback.
    • Fixed initial compile failure (missing CONTROL DSP owner) and capped total validated DSP + diagnostic DSP + output scratch budget.

    Exact-SHA evidence

    Referenced professional workflows and engineering adaptation

    Remaining P0 (do not mislabel as DONE)

    1. A calibrated speaker profile + true peak/RMS/thermal/excursion-informed protection (measured DAC voltage/amplifier gain/manufacturer limits), limiter telemetry and bounded latency. Current hard digital ceiling is not speaker protection.
    2. Single-authority ARM → ACK → actual callback → LIVE hardware output publication after protection and port identity verification; collision checks, auto-disarm, stale graph/port and leak-prevention tests, verified dummy-load hardware FAT.
    3. Complete versioned .smix LMS session persistence/reopen fail-closed, profile/device rebinding, per-output RT meters/ACK to GUI and full operator workflow qualification.
    4. Bound or explicitly opt-in to diagnostic shadow processing duration/CPU before production release; no permanent unlimited diagnostic overhead. Finish native Windows qualification for M8 exact SHA.

    No hardware LIVE or calibrated-protection claims.

  9. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M9 P0 progress checkpoint: exact PR #635 head 6321f4fd1daa53437f7a82e86667ee949835573a, main reconciled at f31f4c2671a0812163b24d05c813e35ca766890e via merge ee9450923ef115e1f79e27580b90e368065e5d5b (0 behind). Implemented actual CONTROL-prepared voltage/peak/RMS DSP, per-driver safety HPF, gain-reduction telemetry, isolated disarmed meter-only callback PCM, trusted profile native admission and hard-fail GUI ceiling/readback parity. Physical ASIO output ARM remains DENIED. Six LMS portable PCM/ACK/binder fixtures PASS + Fast Native https://github.com/masarray/arsonmix/actions/runs/43438914719 and Fast UI https://github.com/masarray/arsonmix/actions/runs/43438914725. M8 prior full Windows 99/99 PASS https://github.com/masarray/arsonmix/actions/runs/43422637401; exact M9 Windows qualified request is pending. Next blockers: prove calibration with measured DAC/amp and driver manufacturer limits, add safe sole-authority physical ARM/speaker outputs with native actual-callback ACK and dummy-load FAT, add .smix LMS disconnected persistence/rebind and clear per-output protection UX; no production LIVE or physical safety claims.

  10. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M10 resolved prior real red CI causes (SonKu Windows stack SEGFAULT, Chromium PFL semantic mismatch, GCC15 restoreExact redundant vector and one-ULP Harmony mono parity) without disabling -Werror/assertions or changing approved GUI. Added CONTROL-verified 2-second ceiling for isolated LMS post-master DSP capture with AUDIO no-heap and no physical writes, full native per-output readback. Exact current source 7e181af8b49507bb8dc079c05229e45ab47105e0 merged latest main=4a6e47e2905a4aa995f879045a907abee8b1aae5 (0 behind). Fast native + SonKu 1MiB stack https://github.com/masarray/arsonmix/actions/runs/43519984951, Harmony https://github.com/masarray/arsonmix/actions/runs/43519985052, Logical https://github.com/masarray/arsonmix/actions/runs/43519984952 and UI https://github.com/masarray/arsonmix/actions/runs/43519984926 PASS. Exact-head full Windows MSVC/CTest and Chromium qualification requested through PR #635 comments; results pending. Full safety/protection next steps https://github.com/masarray/arsonmix/blob/7e181af8b49507bb8dc079c05229e45ab47105e0/docs/LMS_OUTPUT_PROTECTION_FAT_GATE_20261114.md. Keep draft; ASIO physical ARM/LIVE not implemented nor authorized.

  11. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M11 milestone (2026-11-14) — existing ArSonMix .smix native LMS offline design extension

    Application: ArSonMix. Harmonice is exclusively the built-in vocal-harmony FX in the channel FX Rack, unrelated to the LMS product name.

    Exact source ba18ae67b07b919fdc7a7f6c6fe29a987a252eec in draft PR #635: added engine/lms/LmsOfflineDraft.h with versioned M11 arsonmix.lms.offline-draft.v1; appends optional lmsDraft to existing arsonmix.mixer-session.v1 CODEC and real .smix atomic IO, without breaking legacy files. CONTROL-only native LiveDryRoutingLab::acceptNativeSpeakerDraft now detaches every accepted muted draft (including GUI identity aliases), and captureNativeMixerSession includes the disarmed design. restoreNativeMixerSession validates and keeps it CONTROL-only; no ASIO pin mapping, hardware epoch/graph identity, voltage-model profile, speaker unmute or physical writer survives as an LMS authority. Reopen cannot prepare the DSP until new explicit verified physical binding. Native JUCE tests/native_mixer_session_store_test.cpp covers real .smix export/reopen, invalid field injection and rollback; portable tests/lms_offline_session_draft_test.cpp checks unassigned/muted, alias scrubbing, invalid cutoffs/Q, repeatable safe restore. Existing CMake/fast CI include the test. GUI Chromium V7 regression corrected separately: MASTER intentionally MUTE/SOLO (no PFL), ordinary channel MUTE/SOLO/PFL.

    At head ba18ae67b07b919fdc7a7f6c6fe29a987a252eec: Fast Native C++ including new portable M11 fixture, Logical Mixer, UI/JS, vocal Harmony FX DSP, target preset workflows all PASS. Exact-SHA Windows MSVC/CTest and Full Chromium requested once by owner dispatcher (PR comments); pending. Previous M10 Windows 101/101 PASS run https://github.com/masarray/arsonmix/actions/runs/43520227027 but must not be claimed for M11.

    Next: CONTROL-only calibrated measurement provenance and device identity, ensure saved logical speaker name survives device remapping while physical patch requires explicit verified operator selection, native closed-loop acknowledgement for that selection, full dummy-load multi-ASIO hardware FAT. NO speaker physical ARM/LIVE and no public release on this draft.

  12. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M11 milestone (2026-11-14) — ArSonMix native .smix offline speaker design

    Naming: ArSonMix is the mixer. Harmonice is solely the vocal-harmony effect in the FX Rack, never a name for LMS.

    Exact source ba18ae6 (draft PR #635).

    Added engine/lms/LmsOfflineDraft.h. New optional versioned arsonmix.lms.offline-draft.v1 embedded in the existing arsonmix.mixer-session.v1 native .smix schema, with strict allowlist, finite/budget bounds, older-session compatibility and tamper negative tests.

    Native LiveDryRoutingLab accepts speaker plans only after detached offline snapshot validation; successful muted/control-only speaker plan captured in existing .smix workflow. Cold restore retains DSP design but carries no physical ASIO port IDs, graph/device epoch, unmute state or unverified voltage-model authority. It does not prepare/arm speaker hardware.

    Tests: portable tests/lms_offline_session_draft_test.cpp; native JUCE tests/native_mixer_session_store_test.cpp with real file write/read/corruption. CMake CTest + fast native preflight updated. At exact source all cheap gates PASS. M10 Windows 101/101 PASS previously; exact-M11 Windows and full Chromium requested, results PENDING.

    Full Chromium previous red from faulty assertion: MASTER intentionally has 2 MUTE/SOLO keys; regular channels 3 MUTE/SOLO/PFL. Test fixed, no GUI changed.

    Next: independent measurement provenance and physical pin re-approval, explicit remap UX, dummy-load multi-ASIO hardware FAT. NO speaker physical ARM/LIVE or public speaker release.

  13. masarray commented on Nov 13, 2026

    @masarray
    OwnerAuthor

    M12 — DLMS-specific CI and professional port-reassign UX (2026-11-14)

    Correct scope: ArSonMix is the mixer; Digital Loudspeaker Management System (DLMS) is the speaker workspace; Harmonice is solely the vocal harmony FX Rack plugin, not DLMS. Harmonice DSP runs are NOT evidence of DLMS quality. The focused Harmonice workflow now skips feat/lms-* and feat/dlms-* PR branches; generic CMake edits no longer trigger its DSP test. Common Windows native integration is still separate.

    Implemented in exact head 618ac54505e1180b939a464856c0728714bb05c9:

    • New dedicated .github/workflows/dlms-fast.yml, 7 real portable speaker C++ PCM/safety/RT tests + DLMS JS model/native bridge/CI scope. Latest run https://github.com/masarray/arsonmix/actions/runs/43529370770 PASS; full app Fast UI https://github.com/masarray/arsonmix/actions/runs/43529370748 PASS. Vocal Harmony skipped.
    • New reassignVerifiedPorts pure transactional logic: each old logical speaker output mapped to one explicitly selected, unoccupied exact native-ASIO identity; no ordinal remap. Preserves route, EQ, delay, gain, polarity, limiter intent, geometric coordinates; every resulting output forced MUTED. Rejects duplicate pins, dropped routed or tuned outputs, stale device/graph; no partial commit and no stale ACK reused.
    • site/preview-v2/assets/lms-workspace.js now offers Reassign + keep DSP with a compact review, one-to-one port selectors, no silent auto-mapping, and separate native Validate + subsequent full state readback; old explicit reset option remains.
    • Windows MSVC M11 baseline https://github.com/masarray/arsonmix/actions/runs/43524842462 102/102 CTest PASS (full Windows for source ba18ae67, the later M12 sources changed only JS/UI/workflows/docs). Previous full Chromium run failed mixer PAN-center unrelated to speaker processor; actual 5-page DLMS Chromium smoke PASS in that run before unrelated PAN assertion.
    • Pro software UX comparative research and safety states documented in docs/DLMS_PRO_WORKFLOW_SAFETY_M12_20261114.md, based on d&b R1, Lake Controller, Xilica Designer, Powersoft ArmoníaPlus and Q-SYS primary docs.

    Limits: .smix native CODEC includes disconnected speaker design; direct GUI import/read-only visualization of a restored DLMS draft is next. No calibrated physical output ARM/LIVE, no speaker safety certificate, no hardware FAT or public speaker release. Any additional enablement requires measured DAC/amp/driver models, exclusive actual ASIO pin lease, independent test with dummy load and true callback ACK.

    Exact branch: PR #635, draft, mergeable, main 0 commits behind.

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