feat: make Holocron a trustworthy CV research workbench - #525

Open

frgfm wants to merge 19 commits into

mainfrom

codex/v0.3-research-workbench

Commits on Aug 11, 2026

feat(references): add synthetic character training
frgfm
committed
840f4bc
fix(references): simplify character training setup
frgfm
committed
36eed70
fix(detection): repair YOLOv4 training correctness
frgfm
committed
6c77a4a
fix(references): make training CLIs reproducible
frgfm
committed
8c95b09
feat(trainer): add resumable experiment bundles
frgfm
committed
43d6c5a
feat(models): add programmatic model catalog
frgfm
committed
7a7f610
feat(models): add classifier feature interface
frgfm
committed
24f8d42
feat(detection): report standard COCO AP
frgfm
committed
bc5e0f3
feat(references): migrate character training to v0.3 contracts
frgfm
committed
5492751
feat(models): generate provenance model cards
frgfm
committed
03d01a5
feat(release): verify artifacts and modernize ONNX tooling
frgfm
committed
bc3ff0a

Commits on Aug 12, 2026

feat(release): finalize v0.3 trust gates
frgfm
committed
8e78ee7
fix(checkpoints): restrict state loading
frgfm
committed
6473519
docs(agents): add research workbench guidance
frgfm
committed
17230c2
feat(models): add iFormer classifiers
frgfm
committed
ab0654b
refactor(models): simplify catalog discovery
frgfm
committed
b98a968
fix(ci): align release metadata checks
frgfm
committed
9ba5a74
fix(research): align workflows after adversarial review
frgfm
committed
eb5c43f
fix(models): require pinned Hub revisions
frgfm
committed
5381b5c