-
Loader: accept both dtype/torch_dtype kwarg conventions (diffusers version drift)
-
Remove superseded wan_smoke.py (renamed to backend-agnostic smoke_video.py)
-
Rename wan_smoke.py -> smoke_video.py (backend-agnostic)
-
Phase 2: backend-agnostic smoke script (routes small lane, request from renderer policy)
-
Keep Wan 2.1-1.3B selectable in small lane (prefer='wan21_t2v_1_3b')
-
Phase 2: add CogVideoX-2b to small lane (router picks lowest VRAM first)
-
Phase 2: CogVideoX adapter tests (mirrors Wan test suite)
-
Phase 2: CogVideoRenderer (CogVideoX-2b, fp16, fast small lane)
-
Fix: hub_uri always present in artifact (null when not uploaded)
-
Fix axis swap: Wan landscape width 832, height 480
-
Fix axis swap in test assertion (Wan landscape: width 832, height 480)
-
Fix stub: fake pipeline must be callable
-
Fix frame rounding: round UP to 4n+1 (32 frames -> 33)
-
Fix injection detection: explicit flag instead of bound-method identity
-
Fix: torch import only on real generation path (injection-safe)
-
Phase 2: Wan adapter translation + injected-flow tests
-
Phase 2: real Wan T2V smoke script (tiny clip, fixed seed, artifact + evidence pushed)
-
Phase 2: wire Wan 2.1-1.3B into small lane, Wan 2.2 A14B into production lane
-
Phase 2: WanRenderer adapter (lazy load, injectable for tests)
-
Ledger: Phase 1 actual usage
-
Validate assignment on ProductionPlan (fix pydantic serializer warning)
-
Fix dependency test: validate() reports, topo raises
-
Fix lanes init: DEFAULT_REGISTRIES maps kind -> single renderer
-
Fix append test: append model, not dict
-
Fix cache test seeds
-
Clean up fusion test construction
-
Enable validate_assignment so deltas coerce into typed fields
-
Phase 1: end-to-end mock integration test
-
Phase 1: level1 + fusion + arbiter tests
-
Phase 1: graph + scheduler tests
-
Phase 1: ModelRouter tests
-
Phase 1: ProjectState delta/versioning tests
-
Phase 1: evidence schema tests
-
Phase 1: README
-
Phase 1: Evidyne integration doc
-
Phase 1: resource ledger
-
init
-
init
-
init
-
init
-
init
-
init
-
init
-
Phase 1: package init
-
Add final_outputs field
-
Fix bad import: use evaluation.evidence_fusion
-
Phase 1: OutputCompiler stub
-
Phase 1: Evidyne tool layer + optional FastAPI
-
Phase 1: MovieCompiler (hierarchical plan -> DAG)
-
Phase 1: IntentEngine stub (mock mode)
