Paper arXiv v1.0 — reproducibility bundle

Reproducibility bundle for the Sayvant closed-loop QA arXiv paper:

Napier, A., Wiley, J., Heslin, M. Closed-Loop Quality Assurance for Production Clinical AI Documentation. arXiv preprint (2026).

Contents (all under CC BY 4.0):

• ablation_results.json — Table 5 (Ablation) source: per-condition pass rates with per-note pass counts (n=25 cases × 282 checks). Case identifiers anonymized as CASE-001 through CASE-025.
• ablation_omission_breakdown.json — §8 supplementary omission analysis (5-case subsample).
• em-benchmark-bootstrap-ci.json — §6.3(i) post-F17 latency anchor (22.0s, 95% CI [20.7, 23.4]s on n=23 EM cases).
• reproduce_ablation_ci.py — Python 3.11, seed 42, 10,000 bootstrap resamples. Verified to reproduce the exact CIs in Table 5.
• README.md — full bundle documentation, including which categories of material are deliberately withheld.

To reproduce Table 5:

```
git clone https://github.com/sayvant/SQS-Auditor-paper-data.git
cd SQS-Auditor-paper-data
python3 reproduce_ablation_ci.py
```

Output matches Table 5 of the paper exactly (Wilson 68.8 / 80.5 / 58.9%, per-case 94.3 / 95.9 / 91.6%, paired B−C +1.6 pp 95% CI [−0.2, +3.4]).

A Zenodo DOI will be registered prior to peer-review submission.