# Run 04 reconstruction review

The implementation was reviewed independently for physical accounting, source transfer, historical reconciliation and reproducibility before publication.

## Resolved findings

- Transformer core sections now match the sections used in winding EMF calculations in each joint parameter case.
- Paired integer turns preserve the selected nominal voltage ratio within the stated scenario tolerance; tap allowance is separated from nominal turns.
- Both neighboring phases occupy the shared winding window. Radial build, minimum clearances and outer coil extent are screened against the same core section and limb pitch, including replacement designs.
- R513A uses the same explicit material identity in the model and historical comparison.
- SQLite connections close before atomic replacement and temporary-file cleanup on Windows, including rejected-history paths. Tests retain Python 3.10 compatibility.

The final review found no unresolved critical or important defect in the inspected physical accounting. This is a code and engineering-accounting review, not external product verification, thermal/protection qualification or plant certification. Model transfer remains low confidence.

## Important interpretation limits

The historical Run 04 implementation and full report ending were not recovered. The reconstructed model has a new identity, different assumptions and different results. Numerical agreement with an archived table does not validate either model. Historical material-marginal ranges and reconstructed joint parameter cases have different meanings.

Nested elemental lithium and linked ferrous totals must not be added again to their physical hosts or constituents. Budget residuals are constraints, not additional stock. Battery module/cabinet budget endpoints are marginal envelopes and cannot be summed into a joint partition; actual modeled component stocks are separately constrained by the 590 kg whole-cabinet cap.

Civil works and unallocated offsite infrastructure are excluded. Product-specific drawings, installed conductor schedules, thermal and fault studies, selected compositions and internal-fluid inventories remain evidence priorities.

## Verification commands

Final full-suite result: **270 tests passed** on Python 3.13 with a writable workspace temporary directory. The focused reconstruction checks and numerical screens remain engineering-accounting tests, not empirical validation.

```text
python scripts/reproduce_run04.py
python -m unittest discover -s tests
python scripts/reproduce_run03.py --output ../run03-recheck
```

The Run 03 material balance independently reproduced byte for byte, including 1,717 retained historical entities, 217 captured research artifacts and 155 source hashes. Run 04's preservation audit protects 915 original Git blobs, with named code/document exceptions for metadata and SQLite resource cleanup. Its tests compare two fresh output generations under different Python hash seeds and compare them with the committed artifacts.
