# Reproduce the public continuation

This export contains the original notebook, original failed evaluations, the continuation's inputs/code, raw numerical traces and field checkpoints, released physical scores, figures and CSVs. It is a public-reference study, not validated hardware. No field result is inferred from a training loss.

The recorded Python environment is Python 3.12.3, NumPy 2.3.4, SciPy 1.16.3 and Matplotlib 3.10.7. CPU calculations were single-threaded. Native field models reuse the original nonlinear magnetostatic core. No new runtime, platform provider or generic motor predictor is required.

Run in a working copy when regenerating outputs. Keep the original hashes and frozen records intact. File paths below are relative to the unpacked package root.

## Check the export

```bash
sha256sum -c SHA256SUMS
```

Original v1 CSV/figure arithmetic:

```bash
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 reproduce.py
```

This original script does not rerun the new fault physics. Its output retains the original assumed-current scenarios and failed prototype evaluation; the continuation explains their changed interpretation.

## Recompute the new summaries and figures

```bash
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/mass/summarize_direct.py
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/mass/summarize_material_bridge.py
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/fault/summarize.py
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/presentation/make_figures.py
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/presentation/make_material_inverse_figure.py
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 python3 extension/presentation/make_material_maps.py
python3 extension/validation_owner/score_field_contrast.py
```

Path-only export adaptations are recorded with original and exported hashes in `EXPORT_MANIFEST.json`. The physical fixture prediction and its implementation remain byte-identical to the independently frozen versions. Scoring uses three released plot readings; the full source PDF/vector artwork is not included. To inspect extraction independently, obtain the linked author postprint and generate the fifth PDF page's MuPDF trace as `extension/validation_owner/released/fixture-page5.xml`, then run the unchanged extraction script there. This is a source-reading reproduction, not a fresh holdout.

Matplotlib SVG element IDs/metadata may change on regeneration. Numerical tables and image content, not incidental SVG IDs, are the substantive comparison. The exported checksums describe the delivered files.

## Rerun a field–circuit case

Each `extension/fault/runs/<case>/` has exact inputs, a source snapshot, time trace, field exposure array and summary. Most have an execution receipt; the two initial cases have verified exact-source archives. The failed no-load attempt has a nonzero exit record and no successful summary. The corrected attempt is a separate case.

Example using the current corrected solver, in a fresh output directory:

```bash
OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 MKL_NUM_THREADS=1 \
  python3 extension/fault/run_fault.py \
  --arrangement radial_sus_ring_1.0 --thickness-mm 1.5 \
  --br 1.25 --rpm 1000 --pre-at 300 --ncol 8 --cycles 2 \
  --out reproduced/baseline_1000rpm
```

For exact historical source reproduction, invoke that case's `source/run_fault.py` and use its `inputs.json` arguments, changing only the output path. The initial source snapshots intentionally preserve the older residual normalization. Do not overwrite a frozen case or treat a repaired case as if the failure had not occurred.

`--pre-at` is instantaneous phase peak ampere-turn amplitude in the stated sinusoidal prefault state, not line RMS current. `--resistance-20-ohm` is phase resistance at 20°C; `--winding-C` applies the declared copper coefficient. `--end-L-mH` is balanced per-phase end inductance. `--radial-factor` changes radial discretization. All scenarios have prescribed speed, isolated-neutral/star current balance and linear-recoil magnets. Actual winding connection and protection have not been established.

The `matrix_*.json` files specify all bounded scenarios. The launcher snapshots code per case but is only an execution convenience; no extra infrastructure is needed. Numerical verification receipts and source versions define the scope of each accepted or failed check.

## Other model branches

- Direct equal-torque roots: `extension/mass/direct_equal_torque.py`, `direct_promising.py`, `direct_all_five.py`, `direct_material_bridge.py` and `direct_inverse_br.py`, with their matching input files. They write outputs beside the scripts, so use a working copy. They regenerate saved fields and can take minutes under load.
- Simplified axial calculation: `extension/axial/axial_harmonic.py`, with its `SCOPE.md`, `RESULTS.json` and `REPORT.md`. It must not be applied as an empirical correction to the native motor.
- Conducting-ring calculation: see `extension/fault_eddy/REPORT.md` and its frozen `SCOPE.md`. Its exact source, controls, four case outputs and current/field comparisons are included. It is a separate assumed-conductivity/infinite-axial calculation, not a validated finished-ring model.
- Synthetic checks and independent review are under `extension/validation_owner/`. The physical comparison has been released; its values are no longer unseen data.

No original sensor exports, measured actual ring law, physical finished-arc measurements or measured installed-mass result are invented to complete this record. See the notebook for the observation each missing item would resolve.
