M02: executable study, individual reports and direct fields
==========================================================

Start here. This edition preserves the completed M02 calculation and its failed
accuracy screens. It is not a complete mirror of every derived local artifact.
The optional6,736,362byte n12modal checkpoint is retained privately; its exact hash,
purpose and regeneration command are in publication/RETAINED-CHECKPOINT.json.
All39individual records,36direct displacement fields,3four-cutoff frequency tables,
frozen material/reference inputs, source, tests and current figure source/data are
included in the numerical download. No physical experiment or device is implied.

Actual bounded run:3eigensolutions,36direct frequency solves,12nested cutoff curves
on3shared161-point grids.29preflight tests passed. The total through final manifest
was86.578seconds. No additional time-domain integrations or physical trials occurred.
Independent review separately reconstructed the evidence; see the project's review
download for its source/checks. Do not add review calculations to the main matrix.

Main results: all39jobs passed numerical identity checks. First-six mesh frequency
changes are1.733% for n8-to10 and1.013% for n10-to12, meeting the2% threshold. Finest
direct20kHz compliance and global time/volume RMS stress changes are0.147% and0.211%.
The fixed144-mode whole-band mesh changes remain218.5% and155.5%, failing the original
5% screen. The metric is a pointwise complex-compliance difference with a1%-of-peak
denominator floor, not energy efficiency or a fracture gain.

At n12,96-to144modal whole-band change is2.996%, but the largest144-mode displacement
mass-norm error against the12direct references is8.433%, with2of12exceeding5%.
The error screens must be interpreted together. Fixed-cutoff mesh comparisons still
include modal error; direct samples are not continuum or full-band ground truth.
One20kHz check is not validation of a continuous low-frequency band.

Correct portable commands from this directory:
  python -m pip install -r requirements.txt
  python publication/verify_package.py
  python run_m02.py --output fresh-preflight --preflight-only
  python run_m02.py --output fresh-results
  python postprocessing/make_figures.py --results fresh-results --output fresh-figures

The earlier frozen README's figure command omitted the postprocessing/ prefix;
the command above is authoritative. A complete fresh run regenerates the optional
checkpoint at fresh-results/volume/n12-modal-context.npz if all planned jobs finish.
The runner requires64-bit Windows8+ for tested resource limits; the NumPy/SciPy FEM
itself is portable. Do not silently bypass the guard on other systems.

The frozen numerical snapshot lists the actual runtime source dependencies. Plot
and package scripts were added separately after execution; they are included with
their own hashes and are not retrospectively claimed as solve-time dependencies.
Use fresh output folders. Source and input hashes are checked before and after.

First preflight history:28of29tests passed, but a test of Windows peak-memory counter
<=configured cap failed. Allocation denial itself worked. The unresolved counter
semantics and evidence are retained. The corrected test checks configured limits,
actual rejection and sampled current memory, and preserves the contradictory counter
without treating it as admitted memory. This did not change scientific criteria.
The public failure log replaces only the private workspace path; exact original and
transformed hashes are retained in the transformation record. Raw originals stay
private. The first crowded figure is also preserved locally; final figures-v2 only
corrects minor-axis labels and keeps identical numerical inputs.

This is one homogeneous isotropic steel cube with ideal support and prescribed1N
peak force. Damping is assumed. It does not predict hardness, cracks, plasticity,
rock breakup, safe hardware limits, actuator delivery or physical resonance accuracy.
