Skip to content

SonKu P9.9 DRAFT — musical vocal body, adaptive center focus & dimensional instruments; Venue locked - #730

Closed
masarray wants to merge 9 commits into
feat/sonku-p98-signature-cinema-studio-voicing-20261116from
feat/sonku-p98-p99-vocal-depth-20261116
Closed

masarray wants to merge 9 commits into
feat/sonku-p98-signature-cinema-studio-voicing-20261116from
feat/sonku-p98-p99-vocal-depth-20261116

Conversation

@masarray

@masarray masarray commented Nov 16, 2026 •

Copy link
Copy Markdown
Owner

SonKu P9.9 — engineering-complete handoff (thread boundary)

Implementation complete / Windows-native verified. Frozen review snapshot: 61c304e7432955260ff783ff09610cf4200cfd65. This PR remains DRAFT and stacked on SonKu P9.8 #729, not on main. No automatic merge, no public release, and no assertion of acoustic approval by a headphone listener.

Product outcome (only 8D-SonKu)

  • NEW insert defaults to ORBIT STAGE · DEFAULT from the operator screenshot: clockwise moving orbit, 12.0 seconds/revolution, 127° stereo speaker angle, spatial mix 100%, +5.1 dB display (~1.8 linear), LARGE 28.8×28.8m, source 2.7m, Reflections 100%, Reverb 77%, 0° head yaw/elevation. Typed Signature voicing ID1 deliberately selected so native P9.9 actually runs. No silent changes to old .smix / v11/v12 states.
  • Conservative-but-audible, scene-specific vocal body and center articulation plus directional first reflections and real 330Hz room-body coupling in the existing single native KEMAR/room/limiter graph. No AI stem-separation claim, no second PCM engine, extra latency, high-treble shelf, or new runtime dependency.
  • Existing Signature/Studio/Cinema and the separately approved Venue Mega Space keep their identity. P9.7 approved Venue checkpoint remains immutable: fda0eef669e0a59575c489e149662c1bf878c998.

Safety for OTHER products/threads

  • P9.9 vocal processing requires an explicit per-instance SonKu opt-in, not just a project-wide compile flag. Derived/shared-core consumers do not inherit the SonKu audio processing or its Orbit factory default.
  • Protected native derived-product scene admission, control-side unpublished-state adoption, native ACK/readback and publication safety are tested. ImmerSIB's voicing IDs 4/5, scene graph, private state, preset choices and source branch remain owned by [DRAFT — ImmerSIB P9.7 Venue + P9.8 Detail/Impact] Single native SonKu engine, concert GUI #722/[DRAFT] ImmerSIB Concert Front Row XL + warm vocal/room cohesion and bounded makeup #732.
  • Native FX floating window lifetime/ownership fix(native FX): mixer-owned floating editors and reliable FX Rack reopen #731, Guitar FX, Harmonice, mixer transport/session, and other draft PRs are out of scope.
  • NO additional workflow files, global/required workflow gates, Harmonice checks, branch-push watchers, auto-merge, forced rebase, or downstream PR edits. The existing .github/workflows/p1-windows.yml has SonKu branch-name gated P9.9 source/native cases only; the release-default CMake option ARMIX_SONKU_EXPERIMENTAL_P99_VOCAL_DEPTH=OFF. Tests in this PR must not become required for unrelated products. Do not copy this entire workflow or SonKu native headers into other branches.

Exact-head evidence (verified)

  • Windows JUCE x64: run #43904727228 SUCCESS, all 13/13 native CTests passed (including P9.9 default Orbit, P9.9 vocal room and v11/v12 state+ACK).
  • Chromium desktop and narrow/native editor: run #43904727209 SUCCESS. All other relevant GitHub workflows completed success; historical P8.3 physical speaker research gate is intentionally not applicable to this branch.
  • Audio isolation proof: 24/24 full native PCM hashes bit-for-bit identical to P9.8 at 44.1/48 kHz × 0°/180° × Venue/Signature/Studio/Cinema/ROOM OFF/MIX0 when P9.9 is compiled but the branded/core instance is not opted in. Real SonKu instances retain their opted-in musical voicing. Venue/ROOM OFF/MIX0 unchanged by the new SonKu processing.
  • Portable build artifact #14245082679 (8,358,164 bytes). Its historical P9.4 artifact filename is not the product/engine version; embedded provenance points to P9.9 exact commit above. Artifacts may expire; Git commit SHA and these PR records remain permanent provenance.
  • End-to-end reconciliation: audit comment; docs/SONKU_P99_CROSS_THREAD_RECONCILIATION_20261116.md; docs/SONKU_P99_MUSICAL_VOCAL_DEPTH_FAT_20261116.md.

Stop conditions / future owner's scope

SonKu P9.9 engineering work in this thread ends at the frozen commit above. Do not add cross-product gates or churn this stable branch just to achieve a universal green matrix. Completing the whole repository release is a different, explicitly reviewed integration workstream: three-way reconcile SonKu #729/#730 with independent ImmerSIB #722/#732 and native FX owner #731, preserve each product's typed scene IDs and session contracts, rerun a focused combined-native regression, and obtain actual level-matched ASIO/headphone approval before public merge/release.

Disposition: engineering handoff complete; integration and listening acceptance pending; PR intentionally draft / not merged.

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 16, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: 61c304e
Status: ✅  Deploy successful!
Preview URL: https://7f306865.arsonmix.pages.dev
Branch Preview URL: https://checkpoint-sonku-p99-listeni.arsonmix.pages.dev

View logs

Copy link
Copy Markdown
Owner Author

P9.9 first native regression after high-frequency correction — exact commit eed0d8aa6413cf1ab629fdfba54ccfb0a23843a3

PASS source/GUI/PCM in focused P9 run #43870874088 and PASS real Chromium 8D SonKu native editor in UI run #43870874151.

The first candidate failed a real assert at 7.6 kHz (a first-order 760–3350Hz focus band still boosted sibilance). We did not weaken the limit; we changed to two cascaded one-poles on both sides of the band, matching early and late room controls.

Actual native source measurements (44.1 / 48k):

  • CENTER direct 260Hz body 1.09584 / 1.09606 (preserved warmth), 1800Hz 1.13053 / 1.12923 (articulation), 7600Hz 1.01399 / 1.01158 (no harsh shelf).
  • Real first-image SIDE at 1800Hz 1.05958 / 1.05905, center unchanged; late-center return 0.93865 / 0.93942, late-SIDE exactly 1.0.
  • Full single KEMAR + room + P9.8 PCM, 44.1/48k front/rear: Signature 2.05k +7.9–8.6%, Cinema +5.3–5.8%, Studio +2.3–2.7%, 7.5k remains ~0.99–1.02. Venue/ROOM OFF/MIX0 all bit-identical to P9.8, across 12 exact-hash cases. Peak linked guard and finite bounds pass.
  • Anti-phase direct stays exact, 44.1/48k, no room-only ghosts/mono center invention; native JSON v12 preserved.

Not yet a headphone-qualified release. Complete Windows JUCE actual PCM/ACK and packaged GUI job is still required for THIS EXACT SHA (job 133312179007); no merge/release. No inference that frequency measurements prove singing source isolation or human perception. Approved P9.7 Venue checkpoint is unchanged.

Copy link
Copy Markdown
Owner Author

Room-body root cause and corrected full PCM gate — commit 692c0f570d4af9bdc03f3f0edeb0864b19267a35

User hearing observation was correct: P9.7/P9.8/P9.9 vocal center enhancement was injected only into history_ before direct KEMAR FIR; physical room_.process and reverb_.process still received unenhanced x0/x1. Existing P9.9 even reduced some late-mid (720–3400Hz), making enhanced vocal disconnected/dry relative to room.

Actual correction: conservative, center-activity-controlled 330 Hz Q1.4 cascade of two bandpass biquads into both EXISTING room engines. First wall ER send > diffuse late; no added reverb bus, delay, stereo-dry bypass, fake stem separator, ASIO callback allocation, or new RCB/capability. Move center late-cut to 1350–3800Hz to retain 200–650Hz chest/body. Venue scene ID0 / room off / MIX0 keep exact untouched native PCM.

A failed initial 165–670Hz lowpass-difference approach boosted ER input at 320Hz but lost 7.1% of actual 86Hz LF due to phase. Did not loosen bounds. Switched to dual BPF. First gate found/fixed a real gain-state 4-vs-6 length bug before audio ran. The final P9 native source run 43884942225 shows PASS on every 44.1/48k and 0/180° scene with original 86/2050/7500Hz full graph and new 320/2050/7500Hz room probe:

  • Signature 320Hz output ratio 1.168–1.171×; Cinema 1.113–1.141×; Studio 1.076–1.077×, compared to P9.8.
  • Direct 260Hz ~1.096×, 1800Hz ~1.13×, 7600Hz only ~1.012–1.014× in source test.
  • No 86Hz LF regression (same original strict 0.96 lower bound); no high 7.5k gain; 0.98 linked peak guard.
  • Venue, ROOM OFF and MIX 0% identical full PCM hashes for all corresponding rate/direction combinations in BOTH fixtures.
  • Native P9 source, fast contracts, JS scene capability passed. Windows JUCE x64 actual build/ACK/portable package still running job 133361246420 at writing; keep draft / no public release before exact SHA proves compiled and listener A/B.

Note: We cannot truly separate vocalist from centrally panned instruments in final stereo; 'vocal body' means a dynamically selected center band, and vocal warmth is a listener target rather than an objectively verified perception.

Copy link
Copy Markdown
Owner Author

2026-11-16 operator screenshot → genuine new default Orbit (exact head 64e6626e08cbe63c14b1c635da6c55aa43279382)

Operator screenshot is CUSTOM · MANUAL using moving ORBIT clockwise, 12.0s, speaker angle 127°, MIX 100%, output +5.1 dB (=1.8 linear), LARGE 28.8×28.8m, source 2.7m, elevation 0°, HRTF align 0°, HEAD YAW 0°, early REFLECTIONS 100%, REVERB 77%. Screenshot does not expose the hidden v12 sceneVoicing or initial orbit phase, so P9.9 explicitly selects Signature scene ID1 (otherwise P9.9 vocal room coupling is bypassed) and starts at FRONT 0° (deterministic 12s orbit).

Implemented, not just cosmetic: a real C++ SonKuP99DefaultOrbitStage.h source of truth populates SonKuSpatialProcessor only for new P9.9 inserts. Existing old .smix, P9.7 Venue, P9.8 source checkpoint, named music/studio/wide/deep/orbit presets unchanged. New GUI option ORBIT STAGE · DEFAULT is disabled unless the native P9.9 capability flag is read back. Fix sameSettings() omission of native sceneVoicing and hrtfElevationTrimDegrees so a scene-ID-only edit is not silently ignored.

User-requested more audible but bounded Signature DSP: center body/focus, first-image SIDE, late center articulation, and real 330Hz early/late room signals increased modestly. Shared linked peak guard and zero-delta Venue intact. Same v12 native ABI.

Source run #43891308198: PASS. Native 44.1/48k factory orbit azimuth exactly 30° in 1s, peaks 0.288/0.310; original 24-case P9.8/P9.9 Venue PCM lock passes. Dedicated full KEMAR+room 320Hz Signature around 1.198–1.200×, 2.05kHz 1.098–1.106×, 7.5kHz 1.001–1.025× versus P9.8. Source isolated center 260Hz 1.115×, 1800Hz 1.157×, 7600Hz 1.014–1.017×; exact Venue and ROOM OFF/MIX0 unchanged. P9.3 legacy test now asserts original five presets unchanged and P9.9-only sixth properly gated (not a blind change from 5 to 6).

Not yet a release: real Windows x64 JUCE native state ACK + GUI packaging at job 133383748401 must pass for this SHA, plus human level-matched headphone listening. Screenshot footer says NATIVE IMMERSIVE · SCENE VOICING · VENUE LOCKED (P9.8 capability), so it does not prove the operator was hearing P9.9 yet.

Copy link
Copy Markdown
Owner Author

Root-cause repair for red P9.9 Windows CTest (2026-11-16)

Actual failure at P1 Windows run 43891308198, native job 133383748401: CMake built JUCE app and all 13 native test executables; 12/13 passed, but NativeSonKuP90JuceV11AppliedAckAndLegacyRestore asserted at tests/sonku_p90_native_juce_v11_test.cpp:145 !fx->pollCompletion(ack).

Why: the P9.9 operator-requested new Orbit default starts natively in renderMode 9 (old test assumed new instances started in mode 0). The test queued a mode-9-to-mode-9 parameter change (no rendering-mode transition), but incorrectly expected the 12ms two-stage mode fade to delay the ACK. The REAL DSP dispatcher correctly applies same-mode parameter edits in the first PCM callback and ACKs immediately after core_.setSettings(); imposing a fake fade would degrade responsiveness and is architecturally wrong.

Corrected test — commit 94255ba476d1896cbd54904a87f2b9927851a360:

  1. For P9.9 only, test mode9→mode9 parameter-only edit ID800: no ACK before callback, a successful immediate ACK after first callback, and accepted/readback update.
  2. Explicitly stage mode9→legacy mode0 ID799; wait for the true fade completion and ensure quiescence.
  3. Then test the original mode0→mode9 ID801: no early ACK after first PCM block, fade completion only after the real core applies, no premature accepted settings, v12 snapshot and subsequent migration tests preserved.
  4. P9.8/non-P9.9 test behavior unchanged; no DSP, peak, Venue, schema, .smix, packaging or GUI changes. No assertions disabled or CI workflows skipped to achieve green.

The exact-head new P1 Windows run 43898076200 is running. Source preflight, then actual Windows JUCE CTest and artifact upload must be verified before reporting resolved. The approved P9.7 Venue SHA fda0eef669e0a59575c489e149662c1bf878c998 remains untouched.

Copy link
Copy Markdown
Owner Author

✅ P9.9 Windows RED resolved and exact-head verified — 2026-11-16

Commit: 94255ba476d1896cbd54904a87f2b9927851a360 (only ACK regression test semantics changed). Full Windows workflow run #43898076200 SUCCESS:

  • Fast native contracts PASS
  • SonKu P9 exact-source C++ PCM preflight PASS
  • Actual MSVC x64 JUCE ARMixConsole built, 13 / 13 native CTest PASS, 0 failed. Previously red NativeSonKuP90JuceV11AppliedAckAndLegacyRestore now PASS, without removing original deferred mode-switch ACK safety test. Additional same-mode immediate ACK is explicitly tested.
  • NativeSonKuP99AdaptiveVocalDepthVenueLock PASS, NativeSonKuP99NewDefaultOrbitNative PASS, full P89–P98 native and PCM tests PASS.
  • Real portable Windows executable + embedded approved GUI and artifact provenance verified. Artifact ID 14240188073 (currently carries an older P94 artifact label from existing workflow, but the embedded FAT provenance explicitly states P9.9, SHA above) at direct GitHub Actions artifact page.
  • Uploaded ZIP 8,358,642 bytes; GitHub upload SHA256 82301290424043fa78020cdb965598d4fc143164d20ec9b99d25bc55c0e66124.

This resolves the CI red. PR remains DRAFT, not merged or publicly released: software tests are green, actual ASIO/headphone and level-matched listening remain operator FAT. Venue checkpoint fda0eef669e0a59575c489e149662c1bf878c998 and installed existing user .smix untouched.

Copy link
Copy Markdown
Owner Author

Cross-thread reconciliation — additive compatibility guard, NOT a merge

P9.9 head: 61c304e7432955260ff783ff09610cf4200cfd65, 8 changed files / one new commit atop previously Windows-green 94255ba. No writes to main, approved P9.7 Venue SHA fda0eef, ImmerSIB #722/#732, floating-editor #731, factory reset #708 or Guitar FX #668. PR remains draft, base stays P9.8 #729.

Live audit found blocking integration contracts:

Verified at commit above: P1 run #43904727228 has fast native + exact PCM source PASS. 24/24 same-core cases compare FULL P9.8 PCM lines/hashes bit-for-bit while P9.9 is compiled but NOT opted in. Real SonKu opted in remains measurably different; Orbit 12s/127° stays 30°/1s; 44.1/48k, 0/180°, Venue/Studio/Cinema/Signature and off/bypass gates green. Chromium UI run #43904727209 PASS.

Pending at posting: actual x64 JUCE Windows 13 native CTests + portable artifact on this exact SHA. Do not merge/release merely from source PASS. See docs/SONKU_P99_CROSS_THREAD_RECONCILIATION_20261116.md for verified PR topology, ownership and safe integration sequence.

Copy link
Copy Markdown
Owner Author

✅ Reconciliation commit fully Windows-green — exact source verified

PR #730 head: 61c304e7432955260ff783ff09610cf4200cfd65 (one additive commit, 8 files, no unrelated branch writes).

  • P1 Windows #43904727228 SUCCESS. Actual JUCE Windows GUI executable built and verified (not a test-only binary). 13/13 native CTest PASS, 0 failed, including native ACK/legacy state, P9.9 vocal depth, and new 12s Orbit. Portable package provenance explicitly verified source SHA.
  • Chromium UI #43904727209 SUCCESS; remaining relevant workflows success, historic P8.3 rejected research intentionally skipped.
  • Full PCM cross-product isolation: P9.8 vs P9.9-compiled-but-per-instance-unopted = 24/24 exact PCM rows/hashes, 44.1k/48k × front/rear × six scenes. Genuine SonKu opt-in continues to produce audible-band DSP delta in source tests, Orbit reaches exact 30°/sec. P9.7 Venue and OFF/bypass approved locked PCM tests pass.
  • Windows artifact ID 14245082679 (8,358,164 bytes), direct artifact page. Workflow still uses historical P94 artifact prefix; its internal FAT provenance is P9.9 + this exact SHA. Artifact upload SHA256 10be734d3f29b12a4ef10c907398e4579c10491046e7c42929b2075a4b4f6164.

Cross-thread safety verified: main SHA 37985f02... unchanged; approved Venue SHA fda0eef... unchanged; PR #731 native FX editor head 61d64cd... unchanged; PR #732 ImmerSIB head c32345f... unchanged. Their code was neither cherry-picked nor reset. GitHub currently recalculates #732 as mergeable=true to its own base; that does not prove it is compatible with P9.9 (their independent voicing IDs 4/5 and shared CMake/Main changes still require additive reviewed reconciliation). Historical P89/P90 workflow failures on #732 remain a separate backport concern, not caused by this SonKu commit.

Decision: SonKu cross-product regression is now proven by source, full PCM and native Windows. Keep #730 DRAFT, no automatic merge or public release; combined ImmerSIB+SonKu+FX owner build and level-matched headphone/ASIO listening are explicitly not yet verified. Integration checklist in docs/SONKU_P99_CROSS_THREAD_RECONCILIATION_20261116.md.

Copy link
Copy Markdown
Owner Author

✅ Listening-approved SonKu P9.9 integrated into main via selective squash (2026-11-17)

Operator confirmed headphone listening is GOOD and explicitly authorized reconciliation to main.

  • Main: 7b0c2fc1dea359dc98b8807ccd6806ad761ffbd6, merged from PR #733
  • Immutable approved donor: 61c304e7432955260ff783ff09610cf4200cfd65, frozen branch checkpoint/sonku-p99-listening-approved-20261117
  • Donor native JUCE 13/13, PCM + Venue P9.7 lock and Chromium green. Candidate main PR exact SHA cdee12c6f71b35b75ddb9ba3da9a98999b60d119 passed pre-merge main source/UI/Windows preset checks.
  • Main merged Git tree SHA = candidate tree SHA: f385d6002f0844a700a1aa9f61befca5a34986d9. Key DSP, bridge, and GUI blobs are exactly equal to approved donor; all 35 required P9.9 runtime compiler defines match. ARMIX_SONKU_APPROVED_P99_MAIN=ON activates the approved 8D-SonKu plugin in the main ARMixConsole target only; alternate products sharing the core remain per-instance unopted.
  • Zero workflows added or edited, no global product gates, no harmonice CI, no ImmerSIB, Native FX Editor, guitar FX, or unrelated mixer source imported.
  • Main already started its existing post-push Windows/Chromium validation at run 43961314494 and run 43961314335.

Do not merge this old stacked research PR #730 or P9.8 #729 again as a whole into main: #733 selectively integrated their approved SonKu runtime without importing 354 research commits and historical workflow files. Keep the old branch/PR as a provenance reference, not a second deployment/merge mechanism. Future ImmerSIB #732 / Native FX Editor #731 changes must resolve their independent shared-file differences against current main without replacing the approved 8D-SonKu profile.

Copy link
Copy Markdown
Owner Author

Milestone A — ARCHIVED / CLOSED UNMERGED (no code merge or branch deletion). The approved P9.9 SonKu sound was selectively forward-ported via merged #733 using this head as donor; the current main and listening checkpoint are the product authority.

Immutable original head: 61c304e7432955260ff783ff09610cf4200cfd65. Original source branch: feat/sonku-p98-p99-vocal-depth-20261116; original PR target: feat/sonku-p98-signature-cinema-studio-voicing-20261116. Full 47-PR source-head inventory: docs/governance/MILESTONE_A_PR_BRANCH_RECONCILIATION_20261117.md on chore/milestone-a-pr-history-20261117 (planned for main). Closing this review surface is not proof of importing every experimental commit and is not permission to retune released SonKu or ImmerSIB. All branch refs, checkpoints and original code are preserved. Reopening requires a new, precisely scoped owner-approved reason.

@masarray masarray closed this Nov 17, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant