Shipped by Aqua (Self, steward) after Soul PASS-WITH-FINDINGS (cut-samplerate-passthrough.s1 and siblings).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Shipped by Aqua (Self) after Soul PASS (aqua-body:follow_up:soul-pr47-verdict-20261112; two Low findings). UNHEARD.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- test.mjs: a failing test no longer aborts the run, so a mutation shows every test that kills it.
- New test: the noise slider is live (noise 20 vs 100 differ, rasp-band ratio 0.2 at rpm 4800) while noise 100 stays identical to thruster.dsp.
- logic.js dspHash (CRLF folded to LF, then sha-256) is the one normalisation; the page export and analyze.mjs both use it. LF text hashes as before, so old exports stay valid.
- Tests: CRLF == LF hash, LF == plain sha-256, analyze CLI raises no foreign-DSP warning for a CRLF checkout.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Browser check (headless Chromium, 2026-11-10) found the dings page scrolling
to 846px and the playground to 381px at a 360px viewport, and the trial's
Next button staying disabled until both sounds are played while the hint
did not say so. Tables scroll inside themselves on narrow screens, the
playground select and labels may shrink, and the hint says to play both.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
#29 dropped aud()'s match parameter but left the call in plain(): gen_page.py raised TypeError. Equalised
neutrals need no per-element volume, so the call and the unused neu table go.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
finish.py shipped the four neutral clips unmodified (decoded from the Ratatoskr handoff), 4.3-13.7 dB louder
than the senders. build.mjs now renders them (Earcon Workshop warm-bell, default sliders, the handoff's
mention/dm/message/invite = warning/waiting/message/done mapping) and finish.py scales them to the sender LAE
target, which they do not move; they must fit under the true-peak limit there. The generator no longer reads
the handoff dir, so sound-guard drops its --neutral/config and its four known_failures: 0 failing, 0 known.
gen_page loses the now-pointless loudness-match checkbox.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Renders (or takes) the 32 sender and 4 neutral Dings clips and the 24 Earcon Workshop clips and asserts, per clip: one onset per intended
note on the full and high bands, no click/step/DC, no clipping, true peak under a ceiling, settled tails, loudness within a band of the
median. Thresholds live in thresholds.json with a reason each; aqua-ear's own boundaries are pinned. The four neutral fallbacks, 4 to 14 dB
louder than the sender clips, are recorded as expiring known failures. Tests mutate generated clips (v3 knock, clipped tail, click, step,
cut, DC, outlier, echo, late note, ring-over) and require the intended assertion to fail.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The old swap renamed live aside and staging in: between the two renames /aqua/dings/ answered 404 (a
concurrent reader got FileNotFoundError), a kill there left it missing until the next run, and a killed
run leaked .stage-* dirs. Now renameat2(RENAME_EXCHANGE) swaps live and staging in one syscall, a
per-target flock serialises publishers, and the next run deletes stale staging. test_interruption.py kills
the publisher at every filesystem mutation (audit-hook counted), races readers against swaps, and fails
against the old code.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
/aqua/dings/ and /aqua/ear/ are replaced in place (staging dir renamed over the old one, so a failed or
repeated run leaves a whole page); `banners` marks every older per-hash page with a pointer and deletes
nothing. Six stdlib tests: idempotence, replacement, refusal of a non-page, failed publish keeps the old
page, banners added once without touching content, ear index.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Test 6: after note N+1 starts, note N is under -120 dB re its peak, for every Void clip
(fails with the cap reverted: invite note 0 still at -33.4 dB). Page text: the loudness target
moved ~0.005 dB (min over clips, set by void/mention) and Vorbis quantisation is not linear in gain.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Python (numpy, scipy, matplotlib, soundfile). Onsets at 1 ms on a full-band and a >2 kHz channel;
double hit, click/step/DC, clipping/true peak, overlapping tails, abrupt end detectors with stated
assumptions; compare mode. True peak and BS.1770 loudness ported from aqua-feel/metrics.js
(cross-checked against it in a test). 25 synthetic-fixture tests, each detector with a near-miss.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
inspect_clips.py renders spectrogram/envelope PNGs and onset lists for Void and the multi-note
checks for Nibu. gen_page.py now writes the v4 page (unheard) from them. findings.py (v2 vs v3
numbers) had no consumer once the v3 page was published; removed.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Void's soft tick rode every note 95 ms in (190 ms in the mention): a second broadband hit
beside the tone, which is the double hit heard in v3. Removed at the source; a Void note that
something follows now ends by the next onset (invite's note 2 rang under note 3 and made a
second rise). polyphony.test.mjs fails if a Void clip has more onsets than notes (checked in
the full band and above 1.5 kHz, where a tick shows), and checks Nibu notes do not overlap.
Nibu's voice is proposal 3 (8 kHz 5-bit smoothed sine) for all four events; proposals 1 and 2
stay as non-default alternates.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Nobody has heard v3. Each note renders into its own buffer with its own noise stream (a clip is the sum; polyphony.test.mjs). Notes something follows end by the next onset, the last rings (Libby, Kiko). Void's release tick is a soft knock instead of a 0.5 ms click. Kiko is struck glass instead of bowed. Nibu's glided saw is replaced by three no-glide proposals. findings.py and gen_page.py produce the v2-vs-v3 numbers and page.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Eight voices x four notifications on a shared rhythm family, Ogg Vorbis 48 kHz mono,
equal A-weighted exposure, manifest in the dings.ts shape, measurements (true peak, onset,
end-of-clip, pair distance) and a static listening page. Nobody has heard any output.
Imports only the sample rate from tools/earcon-workshop.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The 2026-11-05 aqua-feel edit cut the idle rasp and rattle floors and opened the
rumble low-pass at low load, which narrowed the noise sweep. thruster-trial.dsp gains
a `mix` control (0 = current thruster.dsp, 1 = the v5 mix as cut) and the trial gains
the variant v5noise100: five variants, ten comparisons. thruster-v5-c6561da.dsp is the
old file, a fixture: test.mjs asserts mix=1 equals it sample for sample and mix=0 still
equals patches/thruster/thruster.dsp. analyze.mjs prints mix beside noise%.
Never run in a browser; nothing has been heard.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Record which faustlibraries text each host compiles against (Starfire
2.85.5 has the stalling smooth; the sandbox master has imp(3)). Add
--items smoother and --smoother old|imp3 (same rewritten source on both
sides) and limits fit to the measured deltas; control matching normalises
the leading slash. Report extended; faustlibraries pin proposed, not made.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A Soul pass found the two sides compile against different stdlib text:
the wasm bundle's si.smooth stalls in float32, the sandbox's faustlibraries
master does not. Reword the report, README and render.js; fail on length
mismatch and render errors; check output count; match controls exactly;
default output under the temp dir.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Add FaustNativeRender and native-diff.mjs; render thruster, two sfxr
patches and a tract vowel through both and record the numbers.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The alveolar closure began to open at half the gesture while the velum and the contact held to 0.8, so the oral tract radiated over the last third of the murmur. verify_renders.py now measures the murmur on steady parts against the mean flanking-vowel power; the render test holds both nasals 8 to 16 dB under the vowels.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>