GenoLeWM-FX Borzoi alignment and overlap report¶
Generated by tools.research.fx_borzoi_overlap at 2026-06-11T12:00:00Z.
Parent epic: #266. Contract: #267. Overlap gate: #268. Cache gate: #269.
Decision: go_traitgym_native_borzoi.
The Borzoi rescue overlap audit is an alignment and feasibility artifact only. It is not a model-quality result, clinical result, deployment-readiness result, broad VEP superiority claim, broad Carbon comparison, useful-planning claim, ground-truth biology claim, or exact fipip overlap claim unless the exact fipip table join is actually run.
Reproduce¶
uv run python -m tools.research.fx_borzoi_overlap \
--manifest configs/fx/borzoi_rescue_sources.json \
--output-json docs/research/fx-borzoi-overlap-report.json \
--output-md docs/research/fx-borzoi-overlap-report.md
Machine-readable report: fx-borzoi-overlap-report.json.
Interpretation¶
This report validates the compact TraitGym-native, row-aligned Borzoi score artifact as the first executable rescue substrate. It does not claim exact overlap against the full fipip table unless that optional table join is run.
Artifact Receipts¶
| Artifact | Identity | SHA-256 / checksum | Size |
|---|---|---|---|
traitgym_slice |
songlab/TraitGym:complex_traits_matched_9/test.parquet@1fde19555fe8c0a55b1382bdf4c6f7082209f566 |
sha256:068940684e1e6f735a25a1a4418fb65390628023a96188af7f1a15dd17de7036 |
411336 |
traitgym_borzoi_score |
songlab/TraitGym:complex_traits_matched_9/preds/all/Borzoi_L2_L2.plus.all.parquet@1fde19555fe8c0a55b1382bdf4c6f7082209f566 |
sha256:04ecf61997bad7f84c36021cd2ca2d30d6781b2f56d3a22e65e6742b0cac2a3f |
66797 |
fipip_borzoi_source |
gs://seqnn-share/sniff/borzoi_102_annotation_set/sniff_102_annotations.gz |
md5 KmaNYQ9EEH+5MQvOfCOaZQ==, crc32c DPBpaA== |
18670353280 |
TraitGym-Native Alignment Gate¶
| Check | Value |
|---|---|
| TraitGym rows | 11400 |
| Borzoi score rows | 11400 |
| Usable row-aligned rows | 11400 |
| Minimum usable rows | 10000 |
| Unique variant keys | 11400 |
| Duplicate variant keys | 0 |
| Positive labels | 1140 |
| Negative labels | 10260 |
| Holdout rows | 3390 |
| Train rows | 8010 |
| Finite score rows | 11400 |
| Null score rows | 0 |
Row-order status: assumed_by_traitgym_matched_slice_artifact_layout. The score file has no row ID column, so row alignment is accepted only because the artifact lives under TraitGym's matched-slice preds/all path, uses the same dataset revision, and has the same row count as the public slice.
Optional fipip Exact Join¶
Status: not_run_full_table_not_staged.
The full fipip table was not staged for this run. This report therefore records public fipip object metadata but makes no exact fipip overlap claim.
Blockers¶
No #268 blockers remain for the TraitGym-native row-aligned path.
Recommended Issue Actions¶
| Issue | Action | Reason |
|---|---|---|
| #268 | close-completed |
The TraitGym-native row-aligned Borzoi path passes the 10,000-row alignment gate. |
| #269 | open-next |
Build the manifest-backed cache from the compact row-aligned Borzoi score vector. |