# O0-PHYS-001

Retrospective reproduction of the Delft 2015 Bell test and an exact audit of
the Bell-local bound. Read scientific_record.md for the proof, assumptions,
physical result and limits of the inference to O/0. This package is not a
new physical experiment and does not establish the full O/0 framework.

From this package directory:

```
python -m pip install -r requirements.txt
python src/analyze.py
```

The script downloads the official 152,913-byte archive into memory, verifies
the repository's MD5 checksum and reads its CSV event log. It executes no
downloaded code and does not save or redistribute the source archive.
An existing official download can instead be supplied with:

```
python src/analyze.py --archive PATH/TO/data.zip
```

Expected benchmark: 4,746 raw candidates; 245 selected trials; 196 wins;
S=2.4224998704613703; primary bias-adjusted tail=0.039174643558790086.
The expected published result was known before this analysis, so matching
these values is reproduction, not a novel preregistered finding.

results/summary.json contains source hashes, settings counts, correlations,
statistics, exact-strategy checks and controls. counts.csv preserves aggregate
setting/outcome cells. selected_raw_row_indices.json identifies rows in the
official source. No postselection parameters are optimized in this analysis.

After analysis, regenerate the figure, manifest, hashes and ZIP with:

```
python src/package_report.py
```

The SHA-256 checksum file covers the delivered sources and derived files.
Re-execution updates execution timestamps, so a rerun's summary hash differs
even when every numerical result is identical. Match scientific values as
well as recording source identities.

Source credit: B. Hensen and colleagues, Delft University of Technology,
Nature 526, 682-686 (2015); DOI 10.1038/nature15759. Data repository DOI:
10.4121/uuid:6e19e9b2-4a2d-40b5-8dd3-a660bf3c0a31. The source is governed
by its repository terms, not by a newly assigned license in this package.
