D02 portable supplement - public packaging instructions

Place device-demand/ beside efficiency/ from the prior numerical releases.
The calculator uses Python 3.10+ and NumPy 1.26+. Plotting also uses Matplotlib.
The 13 author unit tests and ideal launch CLI do not require the R07 archive.

From device-demand/:

  python -m unittest test_device_demand -v
  python device_demand.py launch --mass-kg 0.01 --speed 1 --force-cap 0.5 --stroke 0.01 --output launch-001

To reproduce all D02 numerical outputs, first extract the R04, R05, R06 and
R07 core overlays and obtain all 18 separate R07 NPZ files. Preserve their
recorded paths under efficiency/results/iteration07/traces/. Then:

  python device_demand.py study --r07-root ../efficiency --output rerun-001

To regenerate the published figures using the provided frozen summary:

  python plot_demands.py --summary results/D02_summary.json --output figures-rerun-001

Every output argument above must name a new path. Existing evidence is
refused, including an existing empty directory. A numerical rerun can have
different elapsed time and its source inventory can include the optional
plot script added after the original calculation freeze. This does not
alter the original report's four-file source map. Compare numerical fields
and retained input hashes, not the entire JSON file's hash across reruns.

The final published figure directory is figures-v2/. Its separate manifest
binds the plotting source, exact D02 summary and all seven plot/data files.
The first figure layout was retained privately; v2 fixes legend placement.
No numerical input, trajectory, or calculation result changed for the plots.

PUBLIC-FILES.json is an exact allowlist relative to device-demand/. It does
not include itself, local bytecode, prior drafts, parent archives or traces.
The independently prepared reviewer public subset is packaged separately
from output/device-demand-review/public using its own exact allowlist.

D02 performs 1,152 algebraic source/profile screens and derives 36 histories
from 18 existing full traces. It runs zero new trajectory integrations and
zero physical trials. Its four force/stroke profiles are hypothetical. The
observed 0.17 s motion window is not complete launch, guide or catcher travel.
All 36 strikers are still moving at the endpoint. No hardware capability,
fracture, heat limit, repeat-cycle throughput or assembly fit is validated.
