# Reproducibility

The scientific release is already executed; publication packaging did not rerun the full ensembles. Recorded Python/library versions are in ENVIRONMENT.json. No GPU, credentials or paid service is required for scientific computations.

## Inspect without running source

Read the complete manuscript, CLAIMS.json, STATUS.json, RESULTS.md, DATA_DICTIONARY.md and the raw JSON/JSONL files. NPZ snapshots are NumPy numerical arrays. No remote-code dataset loader is provided or required.

## Local quick inspection

From the repository root, with Python 3.10 or later:

```bash
python examples/inspect_release.py
```

This uses only the standard library, checks recorded counts and the matched-control outcomes, and prints the exact ideal reserve threshold. It does not run fabrication or validate a molecular backend.

## Reproduce the scientific layer

Use a separate working copy: runners overwrite their result files and plotting/building scripts overwrite figures/PDFs, invalidating the original release manifest.

```bash
python -m venv .venv
# Activate .venv using the command for your operating system.
python -m pip install -r requirements.txt
python -m unittest discover -s tests -v
python src/run_gauge_revision.py --reps 32
python src/make_gauge_figures.py
```

For the exact recorded scientific dependency versions, use requirements-reproduction.txt instead of requirements.txt. It records the Linux Python 3.12 run, rather than promising compatibility with every interpreter. BLAS implementations can cause small floating-point differences. Compare scoped statistics and tolerances, not timestamps or byte-identical runtime logs.

Preserved v2 commands:

```bash
python src/run_revision.py --reps 32
python src/make_revision_figures.py
```

The consolidated PDF can be rebuilt using `python src/build_pdf.py`; the supplied 48-page PDF was already visually reviewed. See baseline_v1/README.md for v1 execution. Publication validation is separate from the 28 scientific tests.

## Read exported benchmark records

```python
import json
from pathlib import Path
rows = [json.loads(line) for line in Path("data/v3_manufacturing.jsonl").read_text().splitlines()]
assert len(rows) == 512
```

After publication, the same JSONL files are available through the Hub dataset configurations. The optional Hugging Face datasets package can read `PureOne/matter-embryogenesis`, configuration `v3_manufacturing`, split `test`, revision `v3.0.0`. The split is a storage label, not a held-out efficacy study. No training weights exist in this release.

## Immutable reference

Use the verified commit in the publisher's PUBLISH_RECEIPT.json for exact remote provenance, together with PUBLIC_MANIFEST.json for file digests. Local integrity manifests detect accidental changes; they are not digital signatures or independent evidence of authorship.