# Notes

The harness fetched the five public IBDMDB files from the URLs in `data/external.json`; each matched its declared size and SHA-256. `sh code/run` ran in the image built from `env/Dockerfile`, with no network, in about 5 s, and every declared result matched exactly.

Beyond the re-run, I recounted the headline figures with a script of my own (`independent_check.py`), written apart from the bundle's code and reading the same verified files: shared DNA/RNA sample IDs from the HUMAnN3 tables' headers, technical replicates dropped, one metagenomics and one metatranscriptomics metadata record per sample agreeing on participant, diagnosis, and week, and the first sample kept per participant-week. It gives:

- Crohn's cohort: 375 samples from 52 participants (declared 375 and 52).
- PWY-5676 RNA in that cohort: 3 named species strata in the release, 205 samples with positive community RNA, 25 with any named-species RNA (declared 205 and 25).

I did not independently rebuild the triple and event logic; the bundle's own `validate.py` re-reads the six decline episodes' alternative-species RNA from the source rows, and it passed in the container.

Hazard screen: the work points at, rather than copies, de-identified profiles and metadata that their custodians released publicly; the bundle's results carry only public participant and sample codes and pathway abundances, with no clinical values in the event audit. No private data, nothing that helps cause harm, and nothing copied without a right to share it. Answer: none.