SonicPi.js — FX A/B Inspector

Sweep state (2026-05-09 · post capture-wrap fix + AMP=2): mean score 58.4 → 66.9 · HIGH 1 → 7 · LOW 7 → 1 · 13 tier upgrades. Top recoveries (LOW → MID/HIGH): vowel 45→78 ✓, gverb 48→75, hpf 27→53, tanh 41→62, rbpf 18→64, bpf 18→52. New HIGH tier (score≥70 AND MFCC≤180): echo, octaver, panslicer, ping_pong, slicer, vowel, ixi_techno.

fixed tools/capture.ts — the prior in_thread wrap silently dropped FX bus wiring for with_fx + live_loop snippets, capturing dry pass-through audio (4 rhythmic FX collapsed to identical peak 0.666). Fix runs async-by-construction snippets at top level so the engine FX-scope tracker fires. PR also lands (Recorder.ts opus → raw float32) (bus stride-2), MIXER.AMP=2 default, and live-wired Mixer Pre-Amp / Amp Prefs sliders. Remaining LOW (1): nlpf — real upstream WASM ugen anomaly, no longer hidden by capture-tool artifact. INCONCLUSIVE (2): delay, chorus (desktop silent — ). One regression: wobble 75→69 (HIGH→MID) — likely scored HIGH on dry coincidence pre-fix; real FX-applied audio shows timbre divergence.

Companion views: e2e.html — 10 curated compositions A/B; 02_fx_chain recovered 0.15× → 0.75× (chain FX now applies); 07_ambient reverb feedback still over-amplifies 3.22× → 2.46× (upstream WASM ugen).
community.html — 48 community + forum compositions (28 community, 20 in-thread-forum); 46/48 captured, median RMS× 0.78 / peak× 0.94 / MFCC 161. Outliers cluster into two upstream-WASM classes: feedback-FX over-amplification (e.g. 19_ambient_lead 10.8×, 47_fibonacci_zoomies 4.7×) and heavy-filter-chain attenuation (e.g. 43_exploring_tb303 0.09×, 33_dub_reverb 0.10×).
Investigations: raw-lpf.html — 7-pass empirical evidence page ( / / lineage); mono-sample-sp107.html — end-to-end (mono samples played left-channel-only → centered; PR #415), with before/after recordings + per-channel waveforms.

Select an FX from the sidebar.