# Guitar — phase-accurate, sharper pick release

Experimental and unapproved. A plays the unchanged earlier neck-position
physical take. B uses the same complete 248-note score, 162 picks, 86 connected
arrivals, geometry and backing, with sharper pick-coupling withdrawal and a
phase-accurate six-string/shared-bridge stepper. No old guitar PCM is blended
into B or replaced at its existing URL.

The original 1.5 ms pick-spring/dashpot withdrawal suppresses upper harmonics in
held-note probes. A 0.15 ms withdrawal restores much of that spectrum but failed
the original timestep gate with the midpoint solver. Those failures remain local.
The new Gauss backend clears the same temporal gate; making release timing exact
or correcting prescribed velocity alone did not clear it.

The backend solves one structured complex system and recovers both real stages.
Work and dissipation use both stages, not a midpoint approximation. Independent
dense matrices, a separate complete short trajectory, held notes, hammer/pull/
repick, lossless storage and existing-state dependence validate stated numerical
properties. Constant-coefficient free-wave order does not make the whole sampled
instrument fourth-order, physically complete or convincingly guitar-like.

The 16.2 cm / 12 mm sensing aperture and preceding generic electrical circuit
are unchanged. One constant voltage gain matches amplifier-input RMS; one
constant post-amp gain matches guitar playback loudness. The original amp/cabinet,
EQ, room, echo and ending fade exactly reproduce the previous raw amplifier
recording from its previous input. No per-note gain, new melody or bus dynamics.

Important unfinished physics: scheduled bilateral contact, prescribed hand
motion without reacting hand/pick mass, full contact switching-energy accounting,
string bending/polarization, measured hardware and spatial convergence.
Grid refinement also changes node-mass-based finger damping; high-band spatial
differences remain. Bare string/bridge work is not full pick/hand energy proof.

The physical source remains one mono performance before the two cabinet paths.
The separate exact Music II stereo-source benchmark stays available; this model
does not silently claim to reproduce that whole source recipe or solve alto.

The local recipe is `music4/studies/guitar-release-11`: numerical checks, then
`python3 render.py`, `node amplify.mjs`, `python3 export.py`.
Dependencies are the neighboring frozen geometry/observer/electrical source,
score/rhythm and Music II DSP, plus NumPy/SciPy/Numba/SoundFile, Node.js and
ffmpeg. No instrument samples, paid API or agent audio playback is used.