Skip to content

feat(dlms): 5-workspace speaker DSP, safe .smix drafts and explicit ASIO remap (M12) - #635

Draft
masarray wants to merge 117 commits into
mainfrom
feat/lms-five-workspace-logical-20261113
Draft

masarray wants to merge 117 commits into
mainfrom
feat/lms-five-workspace-logical-20261113

Conversation

@masarray

@masarray masarray commented Nov 13, 2026 •

Copy link
Copy Markdown
Owner

Purpose

Deliver an actual ArSonMix Loudspeaker Management offline design milestone in the ONE approved /preview/ renderer, preserving the current PNG-button mixer, WIRE, FX, and native WebView2 composition.

User-facing milestone

  • Speakers toolbar opens/closes a System, Crossover, Matrix, Processing, Alignment workspace in-place; mixer state is untouched.
  • User-selectable mono/stereo and 1–16 crossover ways, logarithmic design-frequency regions and editable ordered split frequencies.
  • Real, reversible, browser-side section→destination matrix draft: one source per destination, source fan-out, no silent fan-in. Demo output roster 2–32 placeholders is explicitly NOT a detected ASIO pin list.
  • Per-output DSP drafts: gain, manual delay, polarity, output gate and limiter fields; muted and uncalibrated by default.
  • Top-view speaker placement and geometric delay based on a reference point (343 m/s), clearly labeled estimated/unapplied.
  • Structural edits (ways/stereo/port identities) require in-app confirmation and invalidate all crosspoints and output DSP drafts. Nothing touches physical hardware.

Architecture/safety

  • scripts/compose_lms.py composes the workspace into the existing deterministic scripts/build_pages.py output, which is also used by the native resource assembler.
  • New CSS is appended ONLY to the existing compiled chrome owner, never a new public renderer or stylesheet stack; original V28 source is unmodified.
  • lms-system-model.js is a pure portable logical model, independent of UI/native/PCM; no JS audio processing, callback, device write or simulated native ACK.
  • This PR does not claim native crossover DSP, actual physical routing, loudspeaker safety certification, or FAT. A second vertical milestone is needed for real JUCE graph/PCM ACK/readback and validated speaker protection.

Checks

  • tests/lms_system_model.test.cjs adds deterministic structure, routing identity, fail-closed, bounds and geometric-alignment fixtures.
  • tests/single_preview_publication_contract_test.py now regenerates LMS composition, asserts exact DOM and asset provenance, and runs the focused model fixtures during inexpensive existing fast UI checks.
  • Awaiting real CI confirmation; no heavy Windows/FAT needed for this GUI+logical draft.

Reference research

Allen & Heath AHM speaker processing: Ways form distinct processed outputs, then each Way is explicitly patched to hardware sockets. Powersoft ArmoníaPlus workflow uses separation of design, configuration and tuning; adapted into five task-focused pages for this project.

Naming: ArSonMix is mixer/application; Harmonice remains only the separate built-in vocal harmony plugin.

Native integration and physical speaker safety follow-up: #636.

M2 native PCM milestone — 13 November 2026

Native DSP is now REAL portable C++ code in engine/lms/PreparedSpeakerSystem.h, co-located on this same product PR rather than split into another speculative architecture PR. CMake adds 2 genuine Windows CTests, Linux fast native runs their exact same source before expensive MSVC.

  • LR4 24dB/oct coefficient banks prepared on CONTROL; up to 16 ways constrained by Nyquist/scratch capacity. Outputs process each section once, then fan-out to explicit exclusive verified native physical pins. No default/guessed device assignment.
  • 8-band output PEQ, per-output gain, fractional-sample delay, mute, and hard digital peak ceiling. THIS IS NOT an acoustically calibrated or true speaker protection limiter; it must not be marketed as such.
  • Immutable candidate prepare, atomic-epoch/revision preflight fields, device pin identity verification, off-RT allocation and rollback on rejection. No heap in audio process.
  • Deterministic PCMs: LR4 low/high frequency response (including both -6dB at crossover with unity sum), multiport stereo, output side separation, alternate physical ordinals, fanout, impulse delay, PEQ boost, invalid Q/frequency, memory bounds, stale device and no-operator-authorisation.
  • Synthetic C++ realtime fixture: stereo 4-way, 8 independent output patches, 3 PEQ bands each, 1500 x 256-frame blocks, no callback heap allocations. The result is not an ASIO performance/FAT certificate.

P0 limitations still open (#636): The kernel is not yet inserted into the existing accepted ChannelId graph and sole hardware callback; browser five-workspace remains a fail-closed offline design preview; output DSP speaker-calibration protection, UI native intent→ACK→snapshot, .smix persistence and real ASIO 2/4/6/8 hardware FAT are not implemented by this milestone. No physical sound may be claimed on the strength of synthetic PCM tests.

Qualification requested: UI and NATIVE once, exact reviewed head 7a0170c0ae509e31f45069028dc967477b8b83cf; fast Linux checks PASS, expensive runs still subject to final outcomes. No duplicate FAT.

M3 UX and physical-output reservation — 2026-11-13

  • Five task-focused LMS pages are built into the single assembled /preview/ GUI. Crossover now graphs actual LR4 electronic magnitude using the same cascaded second-order Butterworth transfer function as the native DSP kernel (annotated design preview at 48 kHz, NOT microphone measurements). Each section is a matrix row; the operator explicitly assigns outputs.
  • Processing is a scoped inspector with EQ / Level & Timing / Protection subtabs. The EQ tab renders an actual 8-band-capable parametric biquad response (not decorative rectangles). A chosen output retains its own PEQ state, gain, polarity, delay and limiter draft.
  • Alignment visualizes speaker-to-reference distances in metres, suggests arrival-time correction relative to the farthest speaker using 343 m/s, and only copies those suggestions into output delay drafts after confirmation. Never automatically changes PCM.
  • Native engine/lms/NativeSpeakerOutputReservation.h is a bounded CONTROL-plane admission prerequisite: MAIN's accepted ChannelId, device epoch, graph revision, sample rate, active output mask and exact port identity must match. Each candidate LMS pin is exclusive of MAIN, MONITOR and installed direct BUS/EXTERNAL pins. Device/graph changes or duplicate/guessed pins fail closed. No live output authorization is granted, even if the UI says limiter ON: the M2 digital ceiling is not real calibrated loudspeaker protection.
  • Extended Node model/interaction tests; new Chromium full-flow smoke at two desktop sizes with screenshots, plus new portable native reservation CTest and fast Linux preflight. Prior flaky V7 test now waits for actual re-rendered hardware key before counting it (no weakening of assertions). Changed-workflow screenshot upload glob YAML has been repaired.
  • Reconciled non-destructively with main 54702c2796b6713dc6a936276b9c5c11b0f97135: preserving latest native PEQ, GUI, WIRE and shared plugin changes. Exact PR HEAD 268c2a945f4b37475e15acdda45c4bcbdb83fc47. Fast UI and Fast Native pass. One exact-SHA UI and one Native CTest milestone requested; results must be checked before merging.

Still unimplemented / NOT release-ready: Native LMS must be attached to the single accepted ChannelId/JUCE mixer graph and physical output publication; add real calibrated speaker limiter/protection plus device-identity ACK/readback and .smix persistence, then full ASIO 2/4/6/8-out hardware FAT. No passive UI draft is physical speaker protection, and this PR remains a draft until those gates are met. Issue #636.

M4: One accepted ChannelId PCM witness + PFL renderer recovery — 13 November 2026

  • The sole DynamicNativeSessionPublisher::render now supports an optional read-only post-routing/post-FX MAIN PCM observer. It executes once after the accepted ChannelId graph processes; it does not alter the MAIN/monitor buffers, change pins, or run extra VST3 graphs. Observe occurs before master hardware gain, and is NOT yet speaker output authority.
  • New native end-to-end lms_native_postfx_main_witness_test.cpp instantiates the REAL accepted native ChannelId roster, routes VOICE→MAIN, doubles its PCM through the native FX dispatch exactly once per audio block, then gives that accepted PCM to M2's preprepared LR4 stereo 4-way kernel in an isolated 12-channel diagnostic memory bank. Verifies 8 split outputs, input-side separation, mute-safe preflight rejection, zero heap allocations over 250 blocks, and unchanged simulated hardware buffers. No physical ASIO writes.
  • The native output polarityInverted and ParametricBand.enabled now match the approved Processing UI's polarity and per-band bypass controls; actual PCM inversion/6dB peaking and no-bypass-leak tests run under Linux and Windows CTest.
  • Traced V7 Chromium failure to a real hardware SVG bug: HTML's data-hw-kind="pfl" was not present in the hardware renderer's COLORS and legend map, while CSS still styled obsolete cue. Fixed mixer-control-keys.js and mixer-control-keys.css to use pfl (click action remains cue); assert PFL/SOLO engraved mask in actual Chromium instead of weakening test.
  • Portable Linux fast gate PASS on exact M4 sources: prepared LR4/PEQ, 1500×256-float 8-output zero-heap stress, ASIO pin reservation, and new accepted post-FX MAIN→8-way crossover witness all pass under -O2 -Werror. Full GUI after previous PFL renderer fix and exact Windows MSVC still require qualification on the reconciled final SHA.
  • Reconciled with current main f76ce45bfb8b1df6d8068cc39224b6bf052a6a3a without overwriting any concurrent mixer changes; final candidate head b8124c737c3e32404f5934e8a8f54c27b60d2d86.

Non-negotiable NOT RELEASED: the observer is not wired to a user-authorized production LMS session; full external output handling, safe calibrated limiting, typed JUCE GUI ACK/readback, .smix, and actual ASIO 2/4/6/8 hardware FAT are still P0. The synthetic test's local output bus must NEVER be mistaken for physical speaker playback. Keep this PR draft until verified.

M5 — Native LMS design transaction with same-graph ACK/readback

  • Actual native code: engine/core/DynamicNativeSessionPublisher.h now exposes control-thread acceptSpeakerDraft(request, realDeviceInventory, acceptedMainChannelId, expectedDesignRevision) with typed SpeakerDraftAck and speakerDraftReadback(epoch, graphRevision). The exact accepted PCM graph's MAIN ChannelId, quantum size, device epoch and graph revision are checked before any acceptance; a verified JUCE output inventory is required (never DEMO pin identities).
  • Transactional: Native speaker reservation + prepared LR4/PEQ/delay compile happen before replacement; stale design revisions, unauthorized source, illegal crossover, physical MAIN/MONITOR/direct-output collisions, unprotected output, or wrong quantum reject without changing prior accepted draft or audio. ACK includes design revision, graph revision, device epoch, scratch bytes and explicit validatedDraft=true, hardwareApplied=false status. Readback contains the entire submitted design and ordered reserved output ordinals.
  • Lifetime / cancellation: Structural PCM graph commit resets prepared speaker design; setExternalOutput revokes it on a successful hardware patch change; unpublish clears it. No second audio renderer, playback callback, or speaker-output write permission has been created.
  • Actual release-mode native test tests/lms_native_draft_ack_readback_test.cpp checks the full accepted ChannelId publisher transaction, exact port identities, revision-CAS behavior, rollback, invalid/unprotected speaker targets, graph revision rollover, and unpublish. It runs under -O2 -Wall -Wextra -Werror -DNDEBUG in cheap Linux fast gate and is registered as Windows CTest LMSNativeSameGraphDesignAckReadbackAndStaleRevocation.
  • Fixed the real V7 Selenium/Playwright assertion: SVG <text> must use Playwright text_content(), not HTML-only inner_text(); retained strict PFL/SOLO legend checks.
  • Exact source candidate: 6e0cc76f46399d2d6dbccf2a38a79b04b6df67e1; fast LMS M2/M3/M4/M5 all PASS in GitHub run 43409922575, UI smoke PASS 43409922571. Base currently f76ce45bfb8b1df6d8068cc39224b6bf052a6a3a; full Windows native and Chromium qualifications are separate from these fast results.

Important: Native typed ACK is not yet connected to the real WebView UI, does not serialize LMS into .smix, and does not physically install speaker DSP after MASTER. The browser correctly remains DESIGN ONLY / OUTPUT DISCONNECTED. No calibrated peak/RMS/voltage/thermal limiter and no real multichannel ASIO FAT. This PR stays draft and MUST NOT be merged as a production speaker controller.

M6 production WebView → native disconnected design, device-mask safety (2026-11-13)

Actual code in reviewed commit 141aa19af739b178b1f02fea213aee6fa1388c5c:

  • engine/lms/NativeSpeakerDeviceBinder.h rejects unverified/stale device, graph, MAIN source, duplicated pin, occupied MAIN/MONITOR/direct route, forged WebView ordinal and any unmuted speaker; it overwrites all browser pin ordinals with the actual accepted output mask from the current device. The tested safe result is a DISCONNECTED design and has no ASIO output authority.
  • LiveDryRoutingLab::inspectNativeSpeakerDevice() builds exactly one trusted ASIO output inventory from JuceActivePortSnapshot and the accepted ChannelId graph/MAIN, including explicit EXTERNAL pin owners. acceptNativeSpeakerDraft() shares this same provenance with M5's prepare+typed ACK; rejected drafts preserve the previously accepted design.
  • LiveAudioEngine::inspectLmsDeviceOffline() and validateLmsDesignOffline() require actual open ASIO, verified DIRECT, non-reconfiguring device, all outputs disarmed, matching callback mask/epoch/quantum and recheck the device after the transaction. Any late change revokes the design; no physical output is armed.
  • src/NativeLmsWebIntent.h decodes strictly bounded typed, revision-fenced JUCE/WebView requests and projects a native-only readback of current real output pins, occupied ordinals, graph/device revisions and accepted muted draft. The GUI cannot supply an accepted speaker-profile calibration.
  • src/Main.cpp registers arMixLmsInspect / arMixLmsValidate JUCE listeners, dispatches CONTROL-only intents, emits arMixLmsInspection and arMixLmsDesignAck. An ACK explicitly says VALIDATED_PENDING_READBACK; there is no optimistic accepted snapshot and no APPLIED claim.
  • The same 5-page site/preview-v2/assets/lms-workspace.js now displays a compact native inspection strip only inside a connected JUCE desktop. Operator explicitly chooses trusted ASIO ports (clearing old crosspoints); Validate is disabled unless the current mask and mute gate match. Browser shows VALIDATED DRAFT only after an independent native inspection/readback matches designRev, graphRev, device epoch, section mapping. Public /preview/ remains offline/design-only. Existing speaker FX/crossover pages remain unchanged in structure.
  • Historical V25/V26 physical-button CSS audit corrected without changing byte-exact ledger: it permits precisely the verified live pfl selector normalization to historical cue and rejects any other CSS edits. This removes the old Chromium v6 rollback false positive without weakening the baseline.

Quality gates: portable native six-test suite PASS on 43413776692; fast UI Node tests/lms_native_bridge.test.cjs three focused public/native/ACK/rejection cases PASS on 43415883810 and current commit 43416772098; Fast Native PASS at 43416772117. JUCE parser projection has dedicated Windows LMSNativeJuceIntentSchemaAndDisconnectedReadback CTest; full native MSVC qualification requested once, https://github.com/masarray/arsonmix/actions/runs/43416938783, in progress at documentation time. No full Windows PASS is claimed before that result.

Blockers before actual speaker output: true calibrated peak/RMS/thermal/excursion safety limiting, post-master PCM-to-physical graph publication with per-pin exclusive ownership and seamless cutover, native .smix offline persistence and startup disarm, real 2/4/6/8 ASIO hardware FAT with dummy load. Draft PR; DO NOT MERGE/ARM. Issue #636.

M8 — actual post-master ChannelId → prepared speaker DSP, isolated realtime diagnostic (2026-11-13)

Exact reviewed head: 6be711496168ebef8c04f02299ab8c43ecfb2d8c (first M8 implementation ce7ee2176dde1b3c00c79d419534b4c29fccdc17; same branch; main=27ffb25802287b81285e61e068ff4e87ae1f6101, 90 ahead/0 behind).

  • Production callback integration, NOT a second engine: LiveDryRoutingLab::process takes the exact successful ChannelId PCM graph revision and runs the existing M2 LR4/section matrix/per-output DSP on post-master gain, master mono and mute slew PCM after the existing graph and FX processed once. M4 pre-master observer was not suitable for output activation.
  • Output separation: No speaker samples are written to physical ASIO buffers. M8 uses only CONTROL-prepared diagnostic-only owned output arrays with explicit ordinal/epoch/graph-revision gating and realtime read leases, plus zero-heap PCM peaks per isolated section. Physical speaker ARM and LIVE remain unavailable.
  • Acceptance/revocation: Existing verified muted draft and reservation still own the design; a new isolated DSP publish occurs transactionally with accepted design CAS. Any graph/device change or explicit revoke disconnects it. Full dual-preparation/output scratch budget is enforced.
  • Tests: Existing native accepted ChannelId VOICE → FX exactly once → 8 private sections, plus 220 blocks of actual post-master scaling, stereo isolation, no physical output mutation, no realtime heap allocation, readback peaks, mute-zero PCM, stale device/graph and revoke. All six LMS portable C++ fixtures PASS under Linux -O2 -DNDEBUG -Wall -Wextra -Werror fast Native. Fast UI PASS. Exact-head MSVC + CTest is in progress, not yet PASS.
  • Still not release-ready: calibrated driver/amp/DAC protection, single-authority hardware ARM/callback ACK, safe physical output activation, per-output GUI LIVE evidence, versioned .smix speaker session recovery, physical ASIO FAT with dummy loads, and bounding/opt-in CPU for the diagnostic witness. No calibrated speaker-protection or physical FAT claims. See issue P0 LMS native engine — crossover sections → matrix → safe multichannel ASIO output #636.

M9 — voltage-aware peak/RMS DSP, disarmed callback witness, strict digital-ceiling parity (2026-11-13)

Current SHA: 6321f4fd1daa53437f7a82e86667ee949835573a on the existing #635 draft branch. Non-destructive reconciliation: two-parent merge ee9450923ef115e1f79e27580b90e368065e5d5b included up-to-date main=f31f4c2671a0812163b24d05c813e35ca766890e and preserved unrelated SonKu/Guitar FX/Vocal Throw/GAIN/SOF mixer changes. At last compare: 0 behind main.

Real DSP implementation:

  • engine/lms/PreparedSpeakerSystem.h accepts CONTROL-only measured-voltage intent per driver: DAC 0 dBFS Vrms, amplifier voltage gain, manufacturer continuous Vrms / peak V, optional 24 dB/oct LR4 safety HPF, RMS detector window and release. Rejects incomplete/nonfinite/out-of-range profiles transactionally. Strictly not hardware calibration proof or true-peak/thermal/excursion certification. No unmute authorization.
  • Prepared sample-peak instantaneous attenuation plus input-RMS/thermal-oriented limiting, release smoothing, existing digital clipping fallback and atomically observed per-output reduction/peak/RMS telemetry. No callback heap/mutex or second callback.
  • src/LiveDryRoutingLab.h now runs isolated diagnostic reference PCM in the REAL accepted ChannelId disarmed meter-only callback, after virtual master gain/mute/mono, while physical ASIO output arrays remain zero. Single FX pass and host device authority retained.
  • engine/core/DynamicNativeSessionPublisher.h exposes per-destination isolated DSP gain reduction/RMS/peak telemetry in CONTROL readback with hardwareApplied=false.
  • M7 parity hole closed: GUI Digital threshold dBFS is now sent in the typed WebView intent and parsed into OutputPatch::ceilingDbfs, returned by native readback, and strictly compared to GUI draft including default -6 dBFS. Hidden native voltage profile fails parity rather than being misrepresented as an identical GUI draft.

Real checks: tests/lms_prepared_speaker_pcm_test.cpp asserts calibrated-math voltage model sample-peak/RMS, gain reduction/recovery, per-pin isolation, nonfinite profile rejection/rollback and LR4 HPF frequency response. Native ChannelId M8 witness now asserts measured-model output reduction and atomic telemetry for exactly one destination while retaining ASIO-buffers-untouched and zero-heap tests. tests/lms_native_draft_ack_readback_test.cpp proves voltage draft remains hardware DISCONNECTED. Native JUCE threshold/hidden-profile assertions and JS negative parity mutations cover GUI bridge.

Not yet release-ready: no ASIO loudspeaker ARM or LIVE, no real verified calibration/microphone measurements, no oversampled true-peak, no independent physical FAT/dummy load, no versioned LMS .smix persistence/reopen, no profile-calibration workflow or speaker-specific HPF approval. Keep PR draft and issue #636 open. Reference: https://support.allen-heath.com/hc/en-gb/articles/40185199750161-Channels ; https://professional.dolby.com/siteassets/products/sls-audio-products/sls-sys-400a/sys400a_processor_settings.pdf

M10 — red-gate root-cause fixes + bounded LMS diagnostic + current-main reconciliation (2026-11-14)

Verified exact head: 7e181af8b49507bb8dc079c05229e45ab47105e0; current main=4a6e47e2905a4aa995f879045a907abee8b1aae5; 0 behind and PR mergeable at last inspection. Two-parent merge 7e181af8 preserved the main alpha.23 landing/README and public-download tests. This is still a DRAFT, no ASIO speaker ARM/LIVE.

Why prior M9 checks were genuinely red (do not conceal):

  • Windows 101 CTest suite: 100/101 PASS; NativeSonKuMeasuredBinauralDirectionalAndChunkParity caused a pre-assert SEGFAULT. Its test used many large fixed-rings SonKuSpatialCore instances on the default Windows test stack. M10 test fixtures now prepare instances on the CONTROL/test heap; production process() remains fixed-memory/no-heap. A real SonKu fixture also runs with ulimit -s 1024 during fast native preflight.
  • Full Chromium: pages_controls_v6_smoke.py incorrectly required data-hw-kind=cue where the approved GUI actually renders data-hw-kind=pfl and the action is cue. Corrected test to assert real PFL semantics without changing GUI art or removing mask/CSS checks.
  • Updated Linux GCC 15 runner: DynamicChannelRoster::restoreExact built a redundant temporary std::vector for budget validation, producing -Werror=free-nonheap-object. withinBudget now accepts a std::span<const ChannelStripDefinition>; no redundant copy, warning not suppressed.
  • GCC 15's independently optimized float accumulation gave a true mono Harmony lead 1 ULP (<1.5e-8) L/R mismatch, failing full bit-exact assertion. HarmonyPitchSynchronous::render now uses the same normalized result only when all contributing L/R source grain samples are exactly equal. Genuine stereo grains continue separate; the test remains bit-exact, no tolerance relaxation.

New M10 native engine optimization: accepted muted speaker drafts get at most two seconds of isolated post-master diagnostic PCM at trusted native sample rate (plus one callback quantum). DynamicNativeSessionPublisher::SpeakerDiagnosticPrepared counts AUDIO-owned frames, automatically stops costly secondary filter processing, exposes capturedFrames, captureFrameLimit, captureComplete and historical peaks/GR via CONTROL readback. Zero heap/physical ASIO writes are verified over all accepted and expired blocks; CONTROL vector allocations are deliberately outside the AUDIO heap fence. No second physical mixer, no output ACL.

Passes at exact head: Fast Native + six portable real LMS C++ PCM fixtures including 1 MiB SonKu stack; Harmony DSP; Logical Mixer; Fast UI/JS parity; targeted native user preset contracts, all PASS.

Expensive exact-SHA gates: requested exactly once via owner dispatcher: /arsonmix-ci native 7e181af8b49507bb8dc079c05229e45ab47105e0 and /arsonmix-ci ui 7e181af8b49507bb8dc079c05229e45ab47105e0. Full Windows 101 CTest and full Chromium MUST finish SUCCESS before claiming M10 hardware-independent qualification; do not reuse prior M8 qualification as M10 PASS.

Professional references and next safe physical plan: M10 calibration and hardware FAT design, based on official Allen & Heath AHM speaker ways/limiters and Q-SYS's manufacturer-verified voltage RMS/peak thresholds. M11 trusted measured profile and .smix disconnected recovery; M12 one-authority physical output transaction + actual callback ACK only after field proof; M13 independent ASIO dummy-load and amplifier/driver calibration FAT. Do not enable speaker LIVE, merge to production, or issue a public speaker release until those gates are evidenced.

M11 — existing .smix disconnected speaker draft

Exact head ba18ae6. ArSonMix is the product; Harmonice is only the built-in vocal harmony FX. Optional versioned M11 LMS speaker design within existing .smix; saved/accepted plans are detached from physical port IDs, graph/device authority, unmute and unverified voltage model. Strict schema whitelist, real .smix roundtrip/tamper negative regression, portable cold-open safety test, same accepted CONTROL graph. All fast checks PASS. Full exact-SHA Windows and Chromium dispatched; results pending. No speaker hardware authorization. Tracking: #636 (comment)

M12 — dedicated DLMS CI + explicit muted ASIO reassignment

At exact head 618ac54505e1180b939a464856c0728714bb05c9, DLMS-focused native PCM and native-web UX PASS (https://github.com/masarray/arsonmix/actions/runs/43529370770), fast shared UI PASS (https://github.com/masarray/arsonmix/actions/runs/43529370748), separate full Windows JUCE/CTest M11 102/102 PASS (https://github.com/masarray/arsonmix/actions/runs/43524842462). M12 adds per-output explicit port map review preserving EQ/delay/gain/polarity/alignment while forcing MUTE, rejecting stale/duplicate/occupied pins, and requiring a new native ACK/readback. Distinct dedicated .github/workflows/dlms-fast.yml contains only seven speaker PCM/safety regressions and speaker UX tests; Harmonice (vocal FX rack) is excluded from DLMS branch runs and must never be reported as DLMS. Pro workflow and unfinished physical FAT: document. Separate previous full Chromium integration run failed a mixer PAN test after PASS 2-viewport DLMS smoke; do not relabel this as DLMS DSP failure nor claim entire Chromium green. All speaker outputs remain physically DISARMED. Issue #636 M12 evidence.

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: 618ac54
Status: ✅  Deploy successful!
Preview URL: https://a81b05a9.arsonmix.pages.dev
Branch Preview URL: https://feat-lms-five-workspace-logi.arsonmix.pages.dev

View logs

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 0859656

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui ba1a84c

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 033a180

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native 6321f4f

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 6321f4f

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native 7e181af

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui 7e181af

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native ba18ae6

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui ba18ae6

@masarray masarray changed the title feat(lms): 5-page crossover/matrix GUI + native JUCE offline design ACK/readback (M6) feat(dlms): 5-workspace speaker DSP, safe .smix drafts and explicit ASIO remap (M12) Nov 13, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant