effects: park Flanger (Phase 3 gate)

[bdbarnett]

Stakes: Phase 3 Flanger stays modulation.Flanger. The Electric Mistress rebuild is not shipped.

Parked, not THROUGH. Gate auditor 2026-09-08, effects/p3-integration. Record: docs/effects-internal/audits/effects-phase3-gate-audit.md. Parent: #31.

Cause: G2 — TAIL / kit.state / allocation unmeasured. G3/G4 — independent pass refutes F3 at 44.1/22.05 kHz (rise 14.624 / 14.832 dB vs ≥15), F6 t90 (identical on NoSlew and wire), F8 audio t60 1.47 s < 2 s; F6 plant silent at constructor defaults. G5/G9 as for the family (board FNV empty; catalogue “unmeasured”).

G6 fits +options FeedbackDelay (5.8/9.8 % vs 9/15).

Trigger: measure the unmeasured Tier 1 cells; regrade or replant so every remaining demonstrated row fires at defaults and cannot be dialled healthy; board digests with a non-class cause; measured cost in the catalogue.

[bdbarnett]

Phase 3 gate audit (2026-09-08, c8b4376) — still PARKED.

G3/G4: independent F3 (44.1/22.05 rise under +15 dB), F6 t90 vacuous (same on NoSlew and wire), F8 audio t60 1.47 s < 2 s. G6 fits 9/15. ADOPTED unchanged.

Record: docs/effects-internal/audits/effects-phase3-gate-audit.md

[bdbarnett]

Phase 3 gate audit at 06dc9a1. Still PARKED. G3/G4: independent F3 three-rate rise (14.624/14.832), F6 t90 vacuous (0.00533 on clean/NoSlew/wire; silent at constructor defaults), F8 declared tail NullBuildGreen / first-pass audio t60 1.47 s < 2 s. G6 9/15 holds. Not adopted.

[bdbarnett]

Phase 3 gate audit at aa40886 (supersedes 06dc9a1): PARKED. G4 not re-run after 07890b4; pre-change independent refuted F3 (44100 rise 14.624 / 22050 14.832). F6 t90 remains vacuous. G6 still fits 9/15. Record: docs/effects-internal/audits/effects-phase3-gate-audit.md.

[bdbarnett]

Deferred to the follow-up program (Brad, 2026-09-08), not fixed and not abandoned.

Three of Flanger's traits passed on a wire — the measurement could not tell the class from nothing — so its demonstrated set was decoration. CircuitPython 10.3.0 ships audiodelays.Flanger, a native swept delay with feedback, and rebuilding vacuous checks on a graph we intend to replace is wasted work.

The follow-up carries the comparison: ours against upstream's on the same traits, probes and boards, answering whether upstream holds what ours claims, what it costs, and whether it moves the class from the audioif tier to stock. That comparison is also the template for asking the same of every class an upstream node could replace.

Staying open, retargeted.