openvax / topiary / 39450347287-1

93%

push

github

Require Python 3.11 and reconcile Isovar audit provenance (#493)

Require Python 3.11+, test through Python 3.14, and release Topiary 5.94.6.
Fork PRs now run the public suite and produce coverage without the private
NetMHC token; trusted runs retain the executable integration tests.

The failing Isovar gate also exposed stale audit provenance. The compact
fixture introduced in #361 retained 55,610 of the original 124,046 alignment
records, but its run metadata still named the original BAM. Both inputs have
now been replayed independently for all 184 entries with published Isovar
1.59.1, Varcode 10.17.3, and PyEnsembl 2.28.3: their outcomes agree exactly.
Complete SAM-record comparisons also prove that CI's openvax-v1 export equals
the compact input, including repeated-record multiplicity.

The refreshed compact-input expectations, run record, and manifests agree.
Compared with the old pin, only five predicted-effect descriptions change
(two also change their container class), as explained by openvax/varcode#603.
RNA counts, protein sequences, mutation intervals, statuses, and filters are
unchanged after normalizing unordered transcript IDs. The PCDH19 context
exception was fixed upstream in openvax/isovar#530, released as Isovar 1.59.1.

New regression checks catch mismatched run/source provenance. Fixture export
rejects a different BAM, inventory, or reference; audit resumption rejects
dependency-version changes. The audit guide documents the source transition
and prediction changes.

Validation:
- `./lint.sh`, additional audit-script lint, and `pip check` pass.
- Both complete 184-entry audit replays agree; all 87 fixture assets verify.
- The provenance regression fails against the previous pin.
- Full repository-local `./test.sh -q --tb=short`: 5,652 passed, no skips,
  94% coverage, including the real NetMHC and optional integration tests.
- Released downstream Vaxrank workflow: 29 passed.
- Wheel and sdist pass `tests/check_distributions.py` and... (continued)

1 of 1 new or added line in 1 file covered. (100.0%)

10270 of 10972 relevant lines covered (93.6%)

0.94 hits per line