RELEASE07 CONTACT-LAW SENSITIVITY — FROZEN INTERNAL PLAN, NOT EXECUTED

Purpose: determine whether the existing timing comparison depends on an
uncalibrated contact-law choice. This remains the finite-energy beam reference,
not a fracture model, calibrated Hertz contact, acoustic laser or device test.
The plan is internally specified before execution, not formally registered by
an outside registry. Six cases were informed by earlier numerical experience;
they are a stress-test subset, not a new statistical sample.

MATRIX AND BUDGET

Keep release06 cases I4-00, I4-01, I4-02, I4-F-error, I4-H-error, I4-H-zero.
Keep both inherited candidate commands and the same case timing perturbation.
Actual fixed-pair input is its result.candidate, not the contextual inherited
adaptive frozen_command field. No estimator, command or parameter is retuned.

Each protocol has20g total moving mass, two10g sources at0.9L, a0.17s main
window and10mJ ideal total mechanical budget. Adaptive uses0.1mJ prescribed
probe plus9.9mJ main at sqrt(.99)m/s; unprobed fixed uses10mJ main at1m/s.
Adaptive retains35s settling and its bounded remainder. The ideal old probe
is not re-excited under each new contact law; real probe transfer could differ.

Three laws x6 cases x2 protocols x2 timesteps =72 primary7-mode rows.
Repeat at5 modes for72 convergence rows:144 planned integrations total.
Steps are5 and2.5microseconds for every law/mode/protocol. Run all7-mode rows
first, then5-mode rows, in the exact case/law/step/protocol order in the JSON.
The two independent protocol states may be vectorized with their own energies.

EXACT CONTACT AND ENERGY DEFINITIONS

Let d=max(striker displacement-local beam displacement,0), and
vplus=max(striker velocity-local beam velocity,0).

  F = K*d^p + C*d^p*vplus
  U = K*d^(p+1)/(p+1)
  dissipation rate = C*d^p*vplus^2

At separation all three are zero. There is no tensile force, adhesion,
position jump, delayed energy injection or velocity reset. Unloading has
repulsive elastic force only. Adding elastic-storage change to the relative
mechanical power loss gives minus the stated nonnegative dissipation rate.
The implementation must change U with p; keeping0.5*k*d^2 for p=1.5 is wrong.

Reference indentation is EXACTLY0.00025m =250micrometres, unchanged for every
case/protocol. It is an uncalibrated normalization point selected before this
run, not a measured indentation or fit to new outcomes. Use each case's old
k[N/m] and a[N*s/m^2]. The illustrative approach speed is1m/s; dissipative
matching below actually holds at the reference indentation for any speed.

1) linear_reference: p=1, K=k, C=a. Exact existing law.
2) hertz_force_matched: p=1.5, K=k/sqrt(d_ref), C=a/sqrt(d_ref).
   At d_ref, elastic and dissipative forces match, hence total force matches.
   Elastic storage is20% lower and elastic tangent stiffness50% higher.
3) hertz_storage_matched: p=1.5, K=1.25*k/sqrt(d_ref), C=a/sqrt(d_ref).
   At d_ref, elastic storage and the dissipative component match.
   Elastic force is25% higher; elastic tangent stiffness87.5% higher.
   The25% is not a statement about total force when damping also acts.

For p=1.5, K units are N/m^1.5 and C units N*s/m^2.5. These are approach-damped
Hertz-exponent surrogates. Real Hertz material parameters require geometry,
moduli and effective radius. No claim of a Hunt-Crossley/restitution fit is made.
A single coefficient cannot simultaneously match force, storage and tangent
stiffness between these exponents; both conventions are therefore retained.

VALIDATION AND ACCEPTANCE

Before the matrix, verify units/coefficient matches (relative1e-12), detachment,
loading/unloading, conservative limit and normalized passive power identity
(tolerance1e-12). Positive indentation must be taken before fractional powers.
Reuse the verified modal basis. Bound contact tangent and velocity damping
across d<=1mm and approach speed<=3m/s; first verify that this speed envelope
covers the passive-energy bound. The exact conservative step-scale screen is
specified in JSON; if it fails, stop and version a new plan rather than silently
changing the declared timesteps.

Every linear row replays the corresponding frozen release06 result: target
energy relative difference<=1e-8, final accounted energy absolute difference
<=1e-10J, and equal feasibility flags. No extra replay integrations are needed.

For accepted rows: finite states; all active sources contact; total energy
residual<=0.5%; each source residual<=0.5% of its own initial energy; contact
indentation<=1mm. Use the conservative whole-span passive modal displacement
bound<=3%L, retaining the11-point sampled value only for compatibility.
Each protocol's coarse/fine target-energy change must be<=2%; its fine-step
5-to7-mode change must be<=2%. A resolved sign flip fails the convergence
screen. Near-equality keeps the existing timestep-sensitivity unresolved
classification. These are internal diagnostics, not physical uncertainty bounds.

Publishable conclusions must preserve absolute energy changes, both matching
conventions, all signs, invalid rows and unresolved cases. A stable ratio does
not validate a law. Peak reversible mode energy is not fracture work.

FULL-STATE TRACES —18 EXACT ROWS, NO EXTRA RUNS

Capture every state, including t=0, for7-mode fine-step rows of I4-00, I4-02
and I4-H-error, all three laws and both protocols. The JSON lists all18 exact
run IDs. At0.17s/2.5microseconds this is68001 samples per trace.
Store float64 time_s[N] and state[N,25] in compressed NPZ, with column/units
metadata in JSON. State contains q7, qdot7, striker positions2, velocities2,
beam dissipation1, contact dissipation2, positive work2 and negative work2.
Use allow_pickle=False when loading. No raw downsampling is permitted.

Reconstruct contact force/storage with that row's p,K,C and revalidate energy,
peaks, contact events and summaries against the trace (rtol1e-10, energy atol
1e-12J). Cumulative losses must be monotone within1e-14J. Plot-only reduction
must preserve extrema/envelopes or select explicit windows; retain raw data.
Trace arrays total about255MB uncompressed. Trace buffers are capped128MiB,
trace artifacts384MiB and total experiment artifacts512MiB. JSON distinguishes
integration completion from trace completion and every trace has a SHA-256.

RUNTIME AND DURABLE OUTPUT

Overall deadline:1800seconds, including preflight, integrations and trace
serialization; permit only a final small checkpoint/manifest after stopping.
Release06 took469.416s for72 rows. Twice as many rows plus nonlinear/trace
costs may fit this budget, but completion is not guaranteed. Respect the
controlling sprint stop/cancellation even when this local budget remains.

Atomically save every completed row and required trace before marking it fully
recorded. Preserve solver/domain failures, interrupted and not-run rows, and
trace failures/quota omissions. On timeout write an honest partial summary.
Do not silently retry, replace cases or adjust coefficients/steps. Snapshot
all source and inherited input hashes before/after. Old releases stay unchanged.

MEASUREMENTS THAT WOULD MAKE THE CONTACT COMPARISON PHYSICALLY USEFUL

Measure actual tip/specimen geometry, roughness, orientation, preload and
fixture compliance. Acquire calibrated synchronized force and relative
indentation/velocity through loading, unloading, rebound and detachment over
relevant operating conditions. Compare force pulse, contact duration,
force-displacement work, restitution and specimen response. Fit competing
laws to the same calibration records and predict held-out impacts without
retuning, retaining parameter uncertainty. Adhesion, crushing, plasticity or
rate effects may invalidate both proposed surrogates. Separately calibrated
specimen elasticity/damping and fracture behavior remain necessary for the
original splitting goal.

PRIMARY BASIS

LAMMPS official granular-contact documentation distinguishes Hooke and Hertz
elastic exponents and the geometry/material dependence of physical Hertz
coefficients: https://docs.lammps.org/pair_granular.html (checked2026-10-06).
The energy/passivity equations, matching conventions and project gates above
are this plan's derivations and engineering choices, not a claim to reproduce
a complete LAMMPS implementation or a prescribed physical test standard.

No batch, hardware operation or publication was performed to prepare this plan.

PLAN JSON SHA-256: f62e0eb1dcb9215f9a5523dd333ee687d142c3d6910cb58cf9d1cd027c31f1e6
