{ "schema": "apsp.lite2.training-validation/v1", "date": "2026-10-23", "status": "offline_contracts_verified_torch_cpu_and_company_b200_gates_pending", "namespace": "training.lite2", "training_source_sha256": "fcb06007c9cb97705d3d4ac0adcf22a9fa03dd993d5a3f82ff3de373ae70891f", "collector_source_sha256_unchanged": "0fe3d02d408041ef446cac53e17ee5257346a07ebf6ccd052eb32c4ea7962179", "runtime_lock_sha256_unchanged": "e404558d31cc6befaa79aa693b99650eee3d269db3d8aec00517af1845aff793", "local_suite": { "command": "python -m pytest tests/training tests/lite2 tests/lite/test_source_distribution.py -q --disable-warnings --maxfail=1", "passed": 45, "skipped": 2, "warnings": 4, "errors": 0, "skipped_tests": [ "test_torch_cpu_forward_backward_and_exact_resume", "test_torch_forced_action_actor_gradient_zero" ], "skip_reason": "No existing Torch on local Windows/WSL interpreters; no package installed or remote spend assumed" }, "verified": [ "Core384 checksums, manifest self-binding, exact seed/checkpoint/foreign keys", "Train-only normalization and train-selected fixed baseline; dev/calibration not fit", "Source-group, seed identity, split/status and feature availability guards", "419d paid-score-only actor observation; no unpaid/future/final/inactive/pruned score exposure", "Integer ledger, final charge, completion DP and infeasible action rejection", "Exact fixed schedule fee equals actual ledger; shared bounded grammar cache", "Provisional cost opt-in and measurement-provenance requirements", "Independent finite-difference reference for masked policy gradient", "Mock allocation gate: exact UUID, one visible device, other jobs rejected without environment mutation", "Existing Lite2 and source distribution regressions", "Python compilation and shell syntax", "LITE2_RELEASE_BINDING_PASS for unchanged qualified collector" ], "not_run": [ "Torch CPU training and optimizer/resume tests", "Company B200 CUDA forward/backward/optimizer and deterministic resume", "Training on company data, full multi-seed study or online evaluation" ], "provided_gates": { "torch_cpu": "env/lite2/check_training_cpu.py", "one_b200": "env/lite2/run-training-smoke.sh", "handoff": "docs/lite/COMPANY_LITE2_TRAINING_SMOKE.md" }, "scope": "Offline replay source implementation; no inference/scorer/model/latent/test access and no scientific efficacy claim", "runtime_rebuilt": false, "modal_launched": false, "company_jobs_modified": false, "existing_receipts_modified": false }