REED STUDY 006 — CONTOUR IN MOTION Build date: 18 October 2026 Playable instrument: https://reed-study.charlesmish.chatgpt.site/contour.html Focused comparisons: https://reed-study.charlesmish.chatgpt.site/contour-listen.html Preserved 005 reference: https://reed-study.charlesmish.chatgpt.site/play.html PURPOSE Charlie preferred Contour, liked Breath-led beginnings, wanted an ending a little longer than Short, and missed some low end. The previous raw Breath range also made it hard to explore: weak settings often did not speak, while strong fresh starts could remain silent. This round develops one playable Contour voice, with one continuous reed-opening axis and a more useful main breath range. It keeps the existing 005 instrument and comparisons intact. The musical question is whether shaping the reed during a phrase feels useful. A static waveform alone cannot answer that. The tests and listening clips therefore include connected notes, moving opening, breath swells, reentries, and endings. These are synthetic model results, not a claim of acoustic clarinet realism or an assertion that the new version sounds better. WHAT TO TRY FIRST 1. Enable sound on the Contour page. Hold C4, or use Hold breath so one hand is free. Keep Breath at Reference and slowly move Reed opening. Does the tone offer useful shades? Does it need a wider or more dramatic range? 2. Try Shape a melody, then play your own short phrase. Opening moves while the same physical state continues; it does not restart the reed. 3. Explore the main Breath slider. Its bottom is a gentle playing pressure, not silence. Release the key to stop. Does this range speak reliably on your device? Does its pitch flexibility feel expressive or distracting? 4. Compare Short, Soft, and Breath on repeated notes. Soft is the new default, with an audible ending closer to Short. It includes a stated output fade. 5. Compare Reference and Warm output, then use the Close listening page with level matching. Listen alone and with the supplied accompaniment. Warm's extra level can otherwise bias the comparison. You do not need to judge lip technique or acoustic realism. Descriptions such as “this change feels useful,” “the note arrives late,” “this ending hangs on,” or “the opening mostly changes brightness” are enough. Saving notes/settings on either page produces a local file; it sends no message or recording. MAIN CONTROLS AND FIXED MECHANICS The voice remains the lumped dynamic Contour reed, using the existing backward-Euler solver and legacy contact law. Flow noise and mouth coupling remain off. The body is still the existing simple feedback tube. This round does not implement a new distributed reed, free reed, throat, or drawn body. Reference New range or behavior Rest opening h0 418.60 micrometers 334.88–502.32 micrometers Closing-pressure Pc 4478.25 Pa Fixed Reed resonance 700 * 4^0.64 Hz Fixed Damping ratio 0.286 Fixed Legacy contact control 0.28 Fixed Main breath gamma 0.8 0.65–0.95 Experimental gamma 0.8 0–2.1, explicitly separate Opening smoothing — 20 ms time constant Output high-pass 950 Hz Reference 950 / Warm 600 Hz The main Breath slider is a normalized position in this measured playing range. It is not a percentage of lung capacity, absolute pressure, or volume. Its center gives gamma 0.8; the endpoints give approximately 2911 and 4254 Pa. The key gate and release policy supply silence. Opening changes h0 while holding Pc, resonance, damping, and contact ratios fixed. It changes gap-flow and reed-pumping coupling. The legacy contact onset and tongue target still scale with h0. This is a useful virtual reed parameter, not a claim that one can physically change a real reed's opening while preserving every dimensional mechanical property. Equivalent mass and stiffness covary in this normalized model. WHY THE WIDER PRESSURE RANGE IS STILL EXPERIMENTAL The reported failure was reproduced as model behavior, without a solver fault: a fresh C4 at gamma 1.1 remained quiet after five seconds, while a note started at 0.8 and then moved to 1.1 could continue sounding. This occurred at both calculation qualities. Main Breath avoids that region; it does not secretly begin at a different pressure, force the oscillator to sound, or repair the underlying branch dependence. Experimental pressure exposes the original wider behavior and warns that fresh notes may fail to start. Calling that behavior “physical hysteresis” would require further evidence. The legacy law still has an estimated static sealing pressure gamma 2.567, above the raw pressure ceiling. Preserving it protects the liked reference sound but leaves the earlier sealing/contact question open. It is not a problem this interface remapping claims to solve. PITCH AND OPENING Opening uses three measured tuning anchors: narrow, reference, and wide. Each contains the thirteen semitones G3–G4 at 44.1, 48, and 96 kHz output, for both calculation qualities. The 78 center entries reuse the exact 005 Contour table; 156 endpoint entries are new. The endpoint measurements use the established reference gesture: gamma 0.8, a fresh 60 ms linear pressure rise, 1.25 seconds total, with the last 250 ms analyzed. Tube delay is interpolated from note and opening. During playback, the opening state is smoothed first, then that actual opening selects the delay, then the engine advances. Note/bend glide is separate. There is no measured- pitch feedback, transposition of rendered audio, or reset on a held opening change. At reference breath the direct-engine check across five opening positions, three notes, three output rates, and both qualities had a maximum error of 0.3675 cents over 90 cases. Breath still changes sounding pitch. In direct moving-opening probes, the largest measured deviations were about 30.6 cents flat at gentle G4 and 22.3 cents sharp at strong G4. Reference-breath motion stayed within 0.35 cents in those sampled windows. The independent actual-controller grid covered 208 fresh starts across all thirteen keyboard notes: every case spoke, with minimum RMS 0.04779 and minimum broad pitch-estimator confidence 0.9733. At reference breath the maximum measured pitch deviation was 2.474 cents. Across the whole main range, the largest deviation was 33.093 cents flat at gentle G4, wide opening, 44.1 kHz/Standard. A gentle wide-opening F#4 at 48 kHz/Higher was 30.564 cents flat. These two cases exceed the initial 30-cent design target. We retained the range with an explicit limitation: breath moved pitch by up to about a third of a semitone in these tests. The target was not silently declared passed. Exact rows are recorded in verification.json. This is a speaking range, not a pitch lock. At C4 and reference breath, narrow/reference/wide openings produced power centroids of 586.2/626.9/679.5 Hz and odd/even harmonic-power ratios of 17.50/15.08/13.16 dB. Overall RMS changed only about 0.45 dB end to end. These level-independent spectral differences show that Opening changes tone; they do not establish how expressive Charlie will find it. BEGINNINGS AND THE SOFT ENDING Breath-led remains the default, using the existing 60 ms smooth pressure rise. Tongued remains available as the earlier small pulse. No hidden preparatory sound or new pressure overshoot was added. Ending Pressure target after note-off Output fade Reset tail Short 70 ms fall 60 ms 500 ms Soft 50 ms hold, then 110 ms fall 120 ms 600 ms Breath 340 ms fall None 850 ms The short pressure hold lets Soft retain its pitch through most of its audible ending; the finishing fade removes the later pressure-driven fall. It is explicitly an output treatment, not proof of improved contact physics. The renderer exposes the physical raw samples separately from finished samples, with finished = raw * output gain. Across G3/C4/G4 at reference settings, the last 10 ms RMS window above 1% of held RMS ended around 60 ms for Short, 110 ms for Soft, and 230–240 ms for Breath. Soft's first 60 ms stayed close to held pitch. Its 80–120 ms window was about 9–15 cents flat at roughly 19 dB below held level. The raw physical tail still falls after the finishing fade. Short-window pitch estimates describe transient periodicity; they are not an audibility verdict. LOW END AND FAIR COMPARISONS Warm changes the first-order output high-pass from 950 to 600 Hz, with a 20 ms transition. It keeps the existing downstream low-pass stages, gain, and protection. The reed, flow, pressure, and bore-feedback states remain identical under the same gesture. This is output voicing, not a remodeled body. There is no automatic gain matching in the playable instrument. The curated dry Warm phrase is about 2.79 dB louder than Reference. On the comparison page, optional matching attenuates the louder selected excerpt; it never boosts a quiet example. Very quiet or excessively mismatched excerpts keep original levels. Full, Attack, Held, and Release intervals are explicit, and matching measures the actual selected interval including any repeat limit. The context pair uses the same dry lead phrase over the same synthetic C-major chord and bass. Warm's lead is attenuated by 0.724955 before adding the backing; Reference's lead is unchanged. The backing and common master gain are identical. Whole-mix matching is disabled so it cannot alter the accompaniment between variants. The accompaniment is a context fixture, not Phosphor, an acoustic recording, or live polyphony in this instrument. The manifest records source hashes, scores, controls, region metrics, WAV hashes, backing construction, and mixing gains. Dry originals retain fixed engine gain. Playback adds only stated 5 ms excerpt-edge fades; original WAV and raw-excerpt downloads do not include these playback treatments. VERIFICATION AND PRACTICAL LIMITS The implementation checks actual sounding starts and broad pitch estimates, not merely finite numbers. Separate engine, controller, interface, listening, and independent checks cover: - Unchanged 005 engine, controller, UI, calibration, and comparison files. - Fresh starts across the keyboard and the chosen breath/opening domain. - Held opening edits without mechanical resets, and return to reference. - Raw/finished tails, exact pre-note silence, and finite release completion. - Exact live-worklet/offline sample agreement at 44.1 and 48 kHz. - Capture timing, request IDs, stale responses, Stop, blur/suspend paths. - Reference/Warm physical-state identity under the same moving score. - All twelve comparison WAVs, interval bounds, source/setting provenance, attenuation policy, and reconstruction of the context mixes. The independent review compared 773,640 reference samples with 005 (maximum sample difference 9.44e-10), checked 100 moving-control pitch windows, and matched 32,000 recorded worklet samples exactly to offline output. It also reconstructed both context WAVs exactly from their published PCM stems and backing. The focused results and accepted pitch-target exceptions are in verification.json. Implementation details live in CONTOUR-ENGINE-006.md, CONTOUR-PERFORMANCE-006.md, and REVIEW-006.md in the source repository. Run npm run test:contour to reproduce the main checks. Rebuilding the comparison dataset uses npm run render:contour. Real browser and iPhone QA were unavailable in this environment. DOM and Web Audio mocks do not prove layout, event delivery, device latency, or real-time audio performance. Higher calculation quality is the default; Standard remains available if a device struggles. The voice is monophonic and currently limited to G3–G4. Numerical convergence, contact realism, and mouth coupling remain separate future work. QUESTIONS FOR CLAUDE / ARGON / GROK Please distinguish measured implementation behavior from physical hypotheses and musical recommendations. The purpose of this build is to assess a specific playable Contour voice before adding more anatomy. 1. Does opening at fixed Pc, resonance, and damping provide a useful first expression axis, given its stated flow/contact and equivalent-mass effects? If it needs a second axis, name one and give a small audible test. 2. Is the remaining breath-induced pitch movement musically useful, or should a future optional note × gamma × opening delay map preserve pitch more closely? Such a map should remain feed-forward and leave an expressive mode available. Do not assume a live pitch servo is required. 3. Soft combines a brief pressure hold and output fade. Does a held tongue stop offer a worthwhile next comparison, with raw and finished tails kept separate, or should we keep this simpler policy if Charlie likes it? 4. Would any proposed contact/sealing change risk destroying the preferred grain? Propose an isolated comparison that preserves the current voice and reports contact duty, closure, onset, and push/recovery behavior. 5. After level matching, does Warm help alone and in the supplied mix? Would a continuous output-weight axis be useful, or just add an unnecessary control? Keep output voicing distinct from changes to acoustic feedback. Charlie’s listening should choose the next branch. A larger body editor or free-reed instrument is still possible, but neither is needed to decide whether this reed can carry a satisfying phrase.