Controls batch 19a: a live level move keeps its own length across a Reposition swap

[tamashi095]

Controls batch 19a: a live level move keeps its own length across a Reposition swap. When a rebuild changes only an automated fader ride's table while a VCA move or a hand move on that fader is still ramping, the move now runs to its end as the uninterrupted engine runs it, and the revision's watermark reports OUTCOME_EXACT. The conservative fallback flag of batch 17 (B17-R14/R15) is deleted.

Closes #1711

What changed

• The stage target did not move; the comparison did (Keep a live level move's own length across a Reposition swap #1711 first item, F1 a3ab8b5). The adoption rule compared the curve with fader_gain, which RememberGain moves while grid events are held. So every heard move landed at the swap. A free stage now compares with the kernel's ramp target (BuiltinFaderBank::ramp_target); a held stage keeps fader_gain.
• The deferred jump (D1, F2 aad5939). At a Reposition adoption a free stage's live ramp is not retargeted. If the target moved, the cell's J jump waits for the move's first free sample (CellState::defer_jump), and a restartable cell holds its grid events until it fires. A free live write and the hold releases clear the deferral. The stretched landing (long_landing, max(J, remaining)) is deleted. Root ruling R19-1: for a moved target with remaining >= J, the old ramp finishes, then the J jump.
• The owed discharge is kept and gated (G4, F3 83113e5, both-host leg 028db34). The self.owed = false line on a free fader write is observable; a gate is red without it.
• The conservative flag is deleted (D2, F4 83113e5). LevelMoveWindow and its TransitionFallback term are gone from crates/control-plane/src/control.rs, with their text in abi.rs, the C header, docs/C_ABI_V1_QUALIFICATION.md and both d.ts files. The Carry a following route's column moves exactly into a six-word successor #1713 route flag and its text stay. Every G1, G2 and R19-1 case asserts OUTCOME_EXACT with no fallback on both hosts.
• Review folds (U1 fold 1 65912a2, 9c164c8; fold 2 6d0a587; unit test 8c2107b). A step of the new table inside a live move's remaining ramp clears the pending deferral, but only when its jump is emitted (otherwise the lane froze until the move's end, or a deferral still owed was dropped). The G2 parity test is a real bit claim against a fresh plan at the fader's node time (the post-insert soft-clip slot has 31 samples of latency).
• Docs: the batch 19 spec check (d86368c; the closed batch 18 specs Render stored mute automation on the strip #1651-Write a following send's route cell from the shared commit #1653 are removed), root's rulings in the specs (e22541d), two decision-record corrections (0bcac82) and the Decide how adoption treats a rounding-only table difference #1716 placeholder (4889236).

Evidence

• audit capi pcm_digest bffc21e857b5e134, unchanged (the audit session has no stored automation). 0 violations, 0 allocations.
• Realtime policy: 168 marked regions (unchanged). No file under crates/engine/src/realtime/ was edited.
• Sizes: no new render state (CellState.deferred already existed). capi_retained 168,722 of 175,296 (main 168,746; -24 B: the B17-R14 Option<LevelMoveWindow> in the control plane's SessionState is gone). Boot growth 19,857,408 bytes, unchanged (pin 20,971,520). simd128 artifact 4,380,022 B (main 4,384,816; -4,794 B, the deleted landing code).
• C header: doc-comment-only change (the B17-R14 sentence removed); no symbol, constant or layout changed; check-capi-abi.sh passes.
• Gates: batch-gates.sh all green on codex/controls-batch-19a at 8c2107b: 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 8c2107b.

Review

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

• U1 (Keep a live level move's own length across a Reposition swap #1711): FAIL (MAJOR-1: a table step inside a deferred move's ramp left a stale deferral, so the lane froze until the move's end while the watermark said EXACT; MINOR-1 an outcome-only test; MINOR-2 a missing gate for a hand mute inside a pending deferral) -> fold 1 -> verification FAIL (MAJOR-A, latent: the fold-1 clear also dropped a deferral when the discontinuity emitted no jump) -> fold 2 -> verification PASS. The verifier also reached the off-grid case in the standard parity session: bit-equal to a fresh plan at the four rates, EXACT.
• Root rulings: R19-1 (above); option C (19a ships Keep a live level move's own length across a Reposition swap #1711 alone; K1, Render stored send gain, mute and matrix automation #1654 and Render stored VCA offset and mute automation #1655 ship in 19b).

Not in this PR

• Render stored send gain, mute and matrix automation #1654 (send rows), Render stored VCA offset and mute automation #1655 (VCA automation) and the shared automation walk K1 ship in batch 19b. The K1 gate for the fold-2 rule (an_unchanged_value_jump_keeps_the_gate_deferral) ships with K1; in 19a the rule is held by an events.rs unit test (a_discontinuity_whose_jump_emits_nothing_keeps_a_pending_deferral, red with the fold-1 rule) and the fader gates.

Placeholders to file

• Decide how adoption treats a rounding-only table difference #1716 "Decide how adoption treats a rounding-only table difference" (.github/ISSUE_SPECS/1716-decide-how-adoption-treats-a-rounding-only-table-difference.md; phase 2). A split table can differ from the old one only in f32 rounding; the cell then jumps over J where a fresh plan never jumps (about 1.7e-4). That is specified D2 behaviour, so EXACT is honest; no gate claims bits for the two-rebuild case.

🤖 Generated with Claude Code

---

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