v0.5.6 — per-walkthrough reproduction notebooks

v0.5.5 — indel + multi-allelic variant prioritization

v0.5.4 — collaborator round 2 (rsID input, LDlink config, AG example …

…IDs)

v0.5.3 — collaborator-audit followups (P1-P3 + ranking sweep)

v0.5.2 — alphagenome equivalence test tolerance fix (closes #81)

Release 0.5.1 — patch

Fixes:
- chrombpnet wide-window predict in chorus base env (P0 from ml007 audit)
- alphagenome_pt 404 + misleading 'ready' message
- chorus backgrounds status missing alphagenome_pt

Pre-existing issues NOT fixed in this patch:
- alphagenome JAX vs PT backend drift (tracked separately)
- chrombpnet integration test panfs timeout (env-specific)

See PR #80 + audits/2026-05-10_v0.5.0_scorched_earth_linux_cuda for details.

Release 0.5.0

Highlights:
- Unified track-rescale across IGV / matplotlib / CoolBox / notebooks
- Symmetric signed rescale for Borzoi RNA / Sei / LentiMPRA
- Uniform DHS-augmented 786-track chrombpnet CDF on HuggingFace
- DHS vocabulary auto-fetch from HF
- ChromBPNet predict_sliding
- Per-layer CDF-sampling guide for new oracle developers

Closes #79.  See PR #78 + audits/2026-05-08, 2026-05-09 for details.

Release 0.4.0: ChromBPNet CDF rebuild + HF mirror consolidation + Alp…

…haGenome PT backend

Release 0.3.0: ChromBPNet HF slim mirror + chrombpnet_nobias default

Chorus 0.2.1

Documentation patch: --all-chrombpnet disk-size claim corrected from
~60 GB to ~100 GB after a user actually ran the opt-in install and
hit a P0 disk-fill issue. Re-measured against ENCODE API: each of
the 42 ChromBPNet ATAC/DNase models is ~1.8 GB on disk (720 MB tarball
+ 1.1 GB extracted, 5 fold ensembles); 42 × 1.8 GB ≈ 76 GB ChromBPNet
weights alone, plus base install ≈ ~100 GB total.

See CHANGELOG.md [0.2.1] for details.