Controls batch 19b: stored VCA offset and mute automation, composed with each member's own curves

[tamashi095]

Controls batch 19b: stored VCA automation renders, on a shared merged automation walk. A VCA's own offset_db and mute rows (#1655) now play on the C ABI and in the browser alike through the VCA's members, in each member's own fader node time. The merged walk K1 composes every curve a lane hears into one state. #1654's route-op machinery ships for its production callers. Send-row automation stays off.

Closes #1655

What changed

• K1, the shared merged walk (crates/automation/src/compose.rs, 202724b; U1/U2 folds). One composed state per output over all the curves it hears (root ruling R19-2 (b)). Events fall at the union of the curves' discontinuities. Coincident events take the longest jump among them (R19-4). Used by the fader stage (Render stored VCA offset and mute automation #1655) and by the following route op (Render stored send gain, mute and matrix automation #1654 machinery).
• Render stored VCA offset and mute automation #1655 VCA offset and mute automation (V0 3d4b70c, V2 4d700ee, V3 b70b899, V4 be4eb15, V5 6a1638b; U3 fold 1 2922d91).
  • A member lane's fader dB is vca_compose_db(own, S + c_1 + ... + c_m): static terms first, then the reaching automated VCA curves in ascending VCA ID (a class B summation order under the standing ruling). A flat VCA ride renders the bits of the same static VCA offset.
  • A member lane's mute is its own curve OR each reaching VCA mute curve OR its other mute terms. A following send of a member follows the VCA mute curves in its tap's node time (V4).
  • A hand move of an automated VCA offset_db writes the entry's offset (live). A hand VCA mute ORs with the curve (live). An edit of a VCA entry, or a change of which member lanes an automated VCA reaches, is a rebuild.
  • The reach-pair memory is charged as bank payload under the existing caps (R19-6). One pair over the cap is refused with a typed diagnostic, not a panic.
  • A session without VCA automation renders the bits it rendered before.
• Render stored send gain, mute and matrix automation #1654 route-op machinery, with send-row automation off (T2 d3f6db6, T3 08c0a25, U2 folds 9381bf0 and fbb252e, B19-R7 fe4a88b, B19-R8 69c156f, B19-R9 51c4fc7; send rows turned off by root ruling B19-R10 in aa8c6ee).
  • The following route op runs on K1. It carries a following route across a plan swap column by column. Its production callers today are source-mute following routes (Follow an automated mute on following sends in render #1652) and VCA-mute following routes (Render stored VCA offset and mute automation #1655 V4).
  • Send rows (route rows 1-6: send gain, mute, matrix) do not render, and the send gain_db takes no hand offset. A session with route-row entries validates, prepares, renders and lists its controls exactly as on main 323433f6a: the entries are inert and there is no new refusal. A new both-host gate holds this (stored_send_row_automation_is_inert_alike_on_both_hosts).
  • Route cells, flag bits 6-11, the gain law and the route value-cursor paths had no production caller left and are deleted.
• The conservative send flag, and its over-report. A rebuild that carries an automated following send (seven words in both plans) while a live send move may still ramp reports CandidateOutcome::TransitionFallback (B19-R7/R8/R9). The text sits beside Carry a following route's column moves exactly into a six-word successor #1713's on all five surfaces: abi.rs, the C header, docs/C_ABI_V1_QUALIFICATION.md and both d.ts files. With send rows off, the flag only over-reports. The B19-R10 check found that every gate change at a carry is bit-equal to the same gate change on one uninterrupted plan. So no mute-only or VCA-mute carry cuts a live send move, but such a rebuild (even an EQ-only one) is still reported as a fallback. Land a removed send row exactly across a live move #1718 deletes the flag when it lands send rows exactly.
• Defined behaviour: the route column's held value. On a route column, a held value follows the column's current free time and never moves earlier (Render stored send gain, mute and matrix automation #1654 T3 rule). A single fader cell instead releases holds at each emitted jump. Unifying the two is phase 2 (Land a removed send row exactly across a live move #1718 "Related").
• Defined behaviour: compose_toward. On a composed fader lane, a record's or an adoption's ramp target composes each curve toward the ramp's completion, each clipped at its own next discontinuity (K1 compose_toward, Compose VCA offsets with stored fader automation #1649 D3). So between the record and another curve's next discontinuity the lane differs from one cell over the summed curve, and that curve's jump makes them equal. A successor plan uses the same rule.
• Merge of main (e0b60a8): batch 19a (Controls batch 19a: a live level move keeps its own length across a Reposition swap #1717) brings Keep a live level move's own length across a Reposition swap #1711 and the Decide how adoption treats a rounding-only table difference #1716 spec. Every conflict took 19b's side, because the 19a commits are patch-equal picks of commits 19b already holds.
• Docs: docs/SESSION_SCHEMA_V1.md documents VCA row rendering and the VCA offset_db offset, and says again that send rows do not render. Spec records: 1654 (status and decision records), 1655 (decision records; U3 MINOR-1 wording 194fe34), Land a removed send row exactly across a live move #1718 (context and gates, 2a39f5e).

Evidence

• LIVE_FOLLOW_ROUTE_OWNER_BYTES: 352 on main, 824 here (x86-64). Reason: a following route now keeps the merged-walk state of both columns inline (FollowCursors: the K1 cursors of the source-mute and VCA-mute curves it hears, the jump rule, the carry flags) and the block's node spans and jump lengths (FollowBlock). The cost applies to each following route only. It is charged under the existing caps.
• capi_retained 168,746 of 175,296 (19a main 168,722; +24 B: the control plane's SessionState keeps the send move window, Option<SendMoveWindow>, where 19a deleted the B17-R14 level-move window; it equals the pre-19a value).
• Boot growth 19,857,408 bytes, unchanged (pin 20,971,520). simd128 artifact 4,480,805 B (main 4,380,022; +100,783 B, +2.3 %: K1, the VCA fader stage and the route op).
• audit capi pcm_digest bffc21e857b5e134, unchanged (the audit session has no stored automation). 0 render errors, 0 allocations.
• Realtime policy: 171 marked regions in 34 files (main 168; the new regions are K1 and route-op render code). No file under crates/engine/src/realtime/ was edited.
• C header: doc-comment-only change (one sentence beside Carry a following route's column moves exactly into a six-word successor #1713's). No symbol, constant or layout changed. check-capi-abi.sh passes.
• Gates: batch-gates.sh all green on codex/controls-batch-19v at 2a39f5e: fmt; bench and realtime policy; host-core clippy default / no-default / all features and the loom spectrum handoff; workspace clippy -D warnings; no-feature release and debug checks; workspace tests (all features); the whole documentation job; builtins, console and effect-runtime fixtures; graph determinism; effect contract; host-core policy; protocol wasm parity; SDK generated and types; C ABI; protocol allocation audit; builtins and builtins-graph trace audits; test-web-audioworklet; check-cross-targets; web build, check-web-audioworklet, SDK headless; CI path routing, test-support CI, browser expected resources, SDK deletions, command reason and kind vocabularies, script reachability; fuzz check.
• Browsers: chromium, firefox and webkit qualification (--check-matrix --self-test-mutations) all pass at 2a39f5e.

Review

A fresh opus-high adversarial reviewer, then a focused verification of each fold.

• U2 (K1 + Render stored send gain, mute and matrix automation #1654): FAIL (four route-column re-time defects) -> fold 1 -> verification FAIL -> fold 2 -> verification FAIL (the owed landing: a further rebuild dropped it; a seek left a stale value). Root ruling B19-R7: split (remove the owed landing, land at once, flag conservatively). Its check FAILed, and so did the B19-R8 check (a broadened flag) and the B19-R9 check (the whole flag term: a move committed while a rebuild is pending, and the landing tail). Root ruling B19-R10: send rows off, machinery kept. B19-R10 check: PASS, no MAJOR. MINOR-1 (the flag only over-reports) is stated above and in Land a removed send row exactly across a live move #1718. MINOR-2 (three stale "cuts a move" comments) is reworded in 2a39f5e. NIT-1 (under the pending-rebuild shape, a removed muting curve opens over the live move's own ramp, as a live record does on one plan) is recorded in Land a removed send row exactly across a live move #1718.
• U3 (Render stored VCA offset and mute automation #1655): FAIL (MAJOR-1: a composed fader lane's hold outlived another curve's step) -> fold 1 -> verification PASS. Its MINOR-1 (the compose_toward wording) is fixed in 194fe34.
• Root rulings R19-1 to R19-6, B19-R7 to B19-R10.

Not in this PR

• Render stored send gain, mute and matrix automation #1654 stays open in phase 2 behind Land a removed send row exactly across a live move #1718. It is the slice that turns send rows on: route cells, flag bits 6-11, the send-gain offset, the gain law and the send-row gates.

Placeholders to file

• Land a removed send row exactly across a live move #1718 "Land a removed send row exactly across a live move" (.github/ISSUE_SPECS/1718-land-a-removed-send-row-exactly-across-a-live-move.md; phase 2). It lands a removed send row (or source-mute curve) exactly across a live send move, and keeps a live send move on any rebuild that carries an automated following send. It deletes the conservative flag. It is amended with the last check's two MAJORs: W1, a live send move committed while a rebuild is pending is cut at adoption and the pending candidate is never marked; W2, the window closes at the record ramp's end while the column's R19-3 landing tail still runs. Render stored send gain, mute and matrix automation #1654 builds on it.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}