E01 exact numerical supplement

This package is self-contained. No prior project archive, workspace path,
hardware connection, API key, CAD kernel or network is required for the models.
Install Python plus NumPy, SciPy and optionally Matplotlib. The recorded run
used Python3.12.14, NumPy2.5.3, SciPy1.18.1 and Matplotlib3.11.2 on Windows.

From the operating-models directory:

  python -m unittest test_operating_models -v
  python run_study.py --output results-rerun-001
  python make_figures_v2.py --results results-rerun-001 --output figures-rerun-001

Output paths must be fresh. The stored results are immutable. Reproduction
does not require downloading existing NPZ traces because the runner computes
new traces. To inspect the original traces, download the40 files listed in
publication/TRACE-INDEX.json and preserve their relative results/ paths.

The main matrix has972 algebraic rows and40 trajectory integrations. The17
author tests include5 additional short analytical/regression integrations;
these are not counted as matrix cases. No physical trial was performed.

E01-PLAN.json froze inputs, cases and numerical screens before integrations.
The report source map binds the original runtime sources. The separately
named make_figures_v2.py was added after results froze; it changes plotting
only. On a reproduction run, that extra source appears in the source map.
Timestamps, runtime, runtime-dependent floating-point details and whole-file
hashes can differ. Compare declared numerical tolerances and input hashes.

Final publication plots are figures-v2/E01_driver_response.png or.svg and
figures-v2/E01_captive_impact.png or.svg. The first figure draft remains in
figures/ and is labeled superseded. Its1ms buffer-force curve misses narrow
peaks visually. V2 instead plots actual stored5us-step target/buffer maxima;
the motion curve remains explicitly downsampled. All plot data and sources
are retained; no dynamics were rerun for this correction.

Read REVIEW-NOTES.txt alongside numerical pass flags. Independent review
found tiny nonmonotonic dense-output cumulative-energy samples. The original
strict audit and subsequent diagnosis are retained by the separate reviewer
package. A successful internal numerical screen is not hardware validation.

PUBLIC-FILES.json is the exact source/input/report/figure allowlist. Files
marked separate_trace are NPZ downloads; the core ZIP excludes them. The
manifest itself is excluded from its own recursive hash inventory. Parent
publication can include this manifest alongside the listed files.
