E01: two catalog driver circuits and a captive gravity-impact reference

Python 3.10+, NumPy 1.26+, SciPy 1.11+; optional Matplotlib for figures.
There is no hardware control, network access, purchase, or physical test.

From this directory:
  python -m unittest test_operating_models -v
  python run_study.py --output results-rerun-001
  python make_figures.py --results results-rerun-001 --output figures-rerun-001

Use new output paths. Never replace frozen outputs to improve an outcome.
The complete internally specified E01-PLAN.json is hash-frozen before tests
that integrate trajectories or the main matrix. This is not an external
preregistration. Sources/inputs are hashed before and after the main study.

ACOUSTIC ELECTROMECHANICAL MODEL

Actual driver-terminal voltage obeys V=Re*i+Le*di/dt+Bl*v. Mechanical motion
obeys Mms*dv/dt=Bl*i-Rms*v-K*x, with dx/dt=v. K=1/Cms+rho*c^2*Sd^2/Vnet.
The sealed rear air volume is a linear, lossless compliance. Leakage, wall
flexibility, port flow, and cavity modes are excluded. Driver Mms includes
the manufacturer's effective air load. Rms=2*pi*Fs*Mms/Qms is a fitted
aggregate loss. No additional piston air mass or radiation damping is added.
This baseline does not isolate radiated sound power or predict SPL/focus.

Stored energy is Le*i^2/2+Mms*v^2/2+K*x^2/2. Signed supply work equals change
in stored energy plus Re*i^2 and Rms*v^2 losses. Positive supplied energy and
returned energy are retained separately. The time-domain integration begins
at rest and its last five cycles are fitted against the independent complex
frequency-domain solution. Catalog nominal values and their rounding are
retained rather than silently adjusted to force consistency.

The 972 steady-state algebraic points are not 972 trajectory integrations.
There are 24 time-domain runs at two numerical resolutions. Each of two
named drivers uses its own source snapshot, not a resized copy of another
driver. The lead volume comes from a CAD geometric displacement proxy;
the half-volume case and both larger-driver boxes are declared sensitivity
assumptions. Any unknown connector/driver displacement remains a build gate.

0.1/0.5/1 Vrms means voltage measured differentially at the driver terminals,
not a PC/WAV volume setting. Amplifier gain, clipping, limiter, output filter,
power-supply behavior, delay and terminal calibration are outside this model.
Xmax is a one-way linear excursion screen. A catalog RMS thermal rating is
not a safe low-frequency drive envelope; no temperature or thermal protection
is calculated without thermal data. Inductance and motor/suspension properties
are constant small-signal approximations, not breakup/large-excursion validation.

CAPTIVE GRAVITY MECHANISM

Coordinate x points down, with target plane x=0. Nominal total moving mass is
12 g and nominal release height85 mm: ideal first-contact energy10.002783 mJ.
The permitted upper mass20 g and release90 mm, plus maximum5 mm descent after
the target plane, yield18.632635 mJ total gravitational work. Actual attached
moving mass must be weighed; CAD volume does not certify this value.

The lower compliant buffer begins at x=2 mm. Its allowed compression is3 mm.
Crossing x=5 mm ends a failed row without a velocity reset or a fictitious
hard-stop force. The model does NOT establish the real shoulder's force or
survival after that event. The upper release/retention domain is x>=-90 mm.
The body is intended to stay captive; physical retention strength, guide
binding, off-axis rotation and attachment loads are unvalidated.

The equations include gravity, a 7-mode300x120x3 mm aluminum reference beam
(E69 GPa,rho2700 kg/m^3), unilateral target contact, guide loss and a passive
lower buffer. Target and buffer use F=k*d+a*d*max(approach_speed,0), stored
energy k*d^2/2, and loss a*d*max(approach_speed,0)^2. Guide force is b*v plus
Fc*tanh(v/vreg); its power loss is nonnegative. This smooth guide model does
not reproduce static stiction. All these contact/loss values are explicit
hypotheses requiring measurement; they are not inferred from printed material.
Actual friction, misalignment or release-pin drag may prevent the12g slider
from dropping under its approximately0.118 N weight. Smooth-drag model release
is not evidence of physical release reliability.

Every state accounts for slider/modal kinetic and elastic energy, gravity,
target/buffer storage, and beam/guide/target/buffer losses. There is no injected
tap energy, active drive, reset, deleted rebound energy, or prescribed stop.
The manual release pin does no work after ideal disengagement. Missing-target
and deliberately soft-buffer cases are part of the frozen matrix.

There are16 main impact integrations:8 cases at10 and5 microsecond steps,
0.60 s maximum. Metrics are checked every RK4 step; retained NPZ traces sample
every50 microseconds and include any terminal boundary-crossing state. Web
traces are further downsampled to about1 ms. Sampled maxima are not certified
continuous-time maxima. End motion and last50 ms near-rest conditions are
reported separately. A finite observation window cannot prove capture/reset.

NUMERICAL SCREENS AND LIMITS

Acoustic amplitude agreement/refinement0.1%, phase0.1 degree, ledger error
1e-5 of final positive supplied energy. Impact ledger error<=0.1% of initial
gravitational energy; target indentation<=1 mm and sampled beam displacement
<=9 mm. Impact coarse/fine force/travel changes<=1% with absolute floors
1e-5 N/1e-6 m; boundary outcome must agree. Failed cases remain in summaries.
These are internal numerical/domain screens, not engineering safety ratings.
A contact/guide/modal dimensional rate screen requires dt*rate<=1 within the
declared contact-travel domain. This conservative scale check is not an RK4
stability proof; ledger and coarse/fine comparisons remain separate evidence.

FDM parts provide housing/fixture geometry; coils, magnets, electronics,
driver, metal guides, fasteners and compliant buffers require actual parts.
This manual one-impact reference does not complete a powered repeated-BPM
exciter. A motor/cam/reset/position-sensor extension needs a separate model
and energy/force/retention checks. No rock/diamond fracture or scalability
from the reference beam to a boulder is predicted here.
