| { |
| "aggregate": { |
| "diagnostics": { |
| "abstention_count": 4, |
| "abstention_rate": 1.0, |
| "edit_distance_mean": 74.75, |
| "predicted_length_mean": 0.0, |
| "runtime_ms_mean": 157750.0, |
| "runtime_ms_total": 631000, |
| "teacher_length_mean": 74.75 |
| }, |
| "primary": { |
| "definition": "reach-ll-onset-with-correct-pre-ll-state", |
| "metric": "reach_ll_rate", |
| "rate": 0.0, |
| "successes": 0, |
| "total": 4 |
| }, |
| "secondary": { |
| "metric": "solved_endpoint_completion_rate", |
| "rate": 0.0, |
| "successes": 0, |
| "total": 4 |
| } |
| }, |
| "benchmark_id": "cubed-core-benchmark-v0", |
| "benchmark_version": "0", |
| "capture_count": 4, |
| "captures": [ |
| { |
| "abstention_reason": "decoder declined to emit a sequence", |
| "capture_id": "2a1a7ff77c7b4d2fa6bb4b09b4aeb851", |
| "deepest_phase_reached": "scramble-start", |
| "edit_distance": 66, |
| "first_unreached_phase": "last-layer-onset", |
| "predicted_length": 0, |
| "reach_ll": false, |
| "reach_ll_hit_step": null, |
| "runtime_ms": 160000, |
| "solved_endpoint_completed": false, |
| "status": "abstained", |
| "teacher_length": 66 |
| }, |
| { |
| "abstention_reason": "decoder declined to emit a sequence", |
| "capture_id": "581da27170754b4083607bd4728ecfde", |
| "deepest_phase_reached": "scramble-start", |
| "edit_distance": 69, |
| "first_unreached_phase": "last-layer-onset", |
| "predicted_length": 0, |
| "reach_ll": false, |
| "reach_ll_hit_step": null, |
| "runtime_ms": 146000, |
| "solved_endpoint_completed": false, |
| "status": "abstained", |
| "teacher_length": 69 |
| }, |
| { |
| "abstention_reason": "decoder declined to emit a sequence", |
| "capture_id": "813fe877c8ec4115ba553d5a4efde21f", |
| "deepest_phase_reached": "scramble-start", |
| "edit_distance": 80, |
| "first_unreached_phase": "last-layer-onset", |
| "predicted_length": 0, |
| "reach_ll": false, |
| "reach_ll_hit_step": null, |
| "runtime_ms": 125000, |
| "solved_endpoint_completed": false, |
| "status": "abstained", |
| "teacher_length": 80 |
| }, |
| { |
| "abstention_reason": "decoder declined to emit a sequence", |
| "capture_id": "c5cc1a9f10b44748b3c7516abcff7d29", |
| "deepest_phase_reached": "scramble-start", |
| "edit_distance": 84, |
| "first_unreached_phase": "last-layer-onset", |
| "predicted_length": 0, |
| "reach_ll": false, |
| "reach_ll_hit_step": null, |
| "runtime_ms": 200000, |
| "solved_endpoint_completed": false, |
| "status": "abstained", |
| "teacher_length": 84 |
| } |
| ], |
| "claim_tier": "interface-boundary-only", |
| "component_diagnostics": { |
| "metrics": [], |
| "scope": "component diagnostics describe tracker/read subproblems and do not establish camera-to-state reconstruction" |
| }, |
| "contract_valid": true, |
| "limitations": [ |
| "This is a development benchmark of complete-solve captures only, not a general dataset and not evidence of generalization.", |
| "All captures come from one solver, one GAN 12 ui smart cube, and one camera setup family.", |
| "Every capture is bound to the single reviewed gtD1 six-color calibration; that calibration was measured under one lighting condition.", |
| "The gt17-gt22 reachLL class and the gtD on-camera-scramble class are not part of this benchmark: at bundle-derivation time no BLE teacher session was bound to them, or their canonical video exceeds this benchmark's 100 MiB video tier.", |
| "Benchmark v0 is a development benchmark, not generalization evidence.", |
| "The evaluator derives the frozen 24 physical whole-cube rotations and requires every held target state to match exactly.", |
| "The prediction schema has no teacher input and the evaluator binds frozen artifacts; the prediction host is not audited for isolation.", |
| "Solved endpoint and edit distance are not substitutes for the primary reach-LL state-trajectory metric." |
| ], |
| "manifest_sha256": "b3e53597678d2c71bc6b787dbb780f4c50018b8336a1d4f0b794e98097c8bbd8", |
| "schema": "cubed-core/benchmark-v0-report", |
| "schema_version": 1, |
| "split": "validation", |
| "split_integrity": { |
| "fixed": true, |
| "group_policy": { |
| "camera_id": { |
| "mode": "disclosed-overlap", |
| "reason": "One solver, one physical cube, and one camera setup family produced the whole corpus, so this dimension necessarily crosses splits. Disclosed rather than hidden: results are within-solver, within-cube, within-setup and are not cross-solver or cross-setup evidence." |
| }, |
| "cube_id": { |
| "mode": "disclosed-overlap", |
| "reason": "One solver, one physical cube, and one camera setup family produced the whole corpus, so this dimension necessarily crosses splits. Disclosed rather than hidden: results are within-solver, within-cube, within-setup and are not cross-solver or cross-setup evidence." |
| }, |
| "session_id": { |
| "mode": "disclosed-overlap", |
| "reason": "One solver, one physical cube, and one camera setup family produced the whole corpus, so this dimension necessarily crosses splits. Disclosed rather than hidden: results are within-solver, within-cube, within-setup and are not cross-solver or cross-setup evidence." |
| }, |
| "setup_id": { |
| "mode": "disclosed-overlap", |
| "reason": "One solver, one physical cube, and one camera setup family produced the whole corpus, so this dimension necessarily crosses splits. Disclosed rather than hidden: results are within-solver, within-cube, within-setup and are not cross-solver or cross-setup evidence." |
| }, |
| "solver_id": { |
| "mode": "disclosed-overlap", |
| "reason": "One solver, one physical cube, and one camera setup family produced the whole corpus, so this dimension necessarily crosses splits. Disclosed rather than hidden: results are within-solver, within-cube, within-setup and are not cross-solver or cross-setup evidence." |
| } |
| }, |
| "observed_cross_split_groups": { |
| "camera_id": [ |
| { |
| "capture_count": 22, |
| "splits": [ |
| "train", |
| "validation", |
| "test" |
| ], |
| "value": "camera-iphone-back-wide-01" |
| } |
| ], |
| "cube_id": [ |
| { |
| "capture_count": 22, |
| "splits": [ |
| "train", |
| "validation", |
| "test" |
| ], |
| "value": "cube-gan12ui-01" |
| } |
| ], |
| "session_id": [ |
| { |
| "capture_count": 22, |
| "splits": [ |
| "train", |
| "validation", |
| "test" |
| ], |
| "value": "session-complete-solves-2026-07" |
| } |
| ], |
| "setup_id": [ |
| { |
| "capture_count": 22, |
| "splits": [ |
| "train", |
| "validation", |
| "test" |
| ], |
| "value": "setup-owner-indoor-01" |
| } |
| ], |
| "solver_id": [ |
| { |
| "capture_count": 22, |
| "splits": [ |
| "train", |
| "validation", |
| "test" |
| ], |
| "value": "solver-owner-01" |
| } |
| ] |
| }, |
| "unit": "capture" |
| }, |
| "system": { |
| "config_sha256": "86fa6c722804f88ce9486b0276032ff0e939af01f6acaba8d4fff1d618f98f5e", |
| "implementation_sha256": "22124598154864394d0da5cd7a74b9b97e89097ee84596525d48fb2c81026c23", |
| "model_artifacts": [ |
| { |
| "name": "alignment-classifier.onnx", |
| "sha256": "d715b66ba3d7b500bf841f22f2ae29273baae03880d473e1b1cc2d3064d1cd6f" |
| }, |
| { |
| "name": "calibration-gan12.json", |
| "sha256": "c259404e7257fa83f4dabf33286dd49cb20696e3216804aa0aaf9010529c48e1" |
| }, |
| { |
| "name": "face-pose.onnx", |
| "sha256": "4e72d686ed0389405695d87f6b7dbccdb3496e87142bd0a7269b0fa3efaea75c" |
| }, |
| { |
| "name": "trust-v1-numpy.npz", |
| "sha256": "3f40e5d5f5f385f8b96547de11b0d92f17d5aa5575302646338e273fb0c5bfd7" |
| } |
| ], |
| "name": "cubed-core-local-camera-v1", |
| "run_receipt_sha256": "9a971a1d0215109ae07fec598745496cc1c4a858578ce12157422aadaab41f0f", |
| "version": "0.2.0" |
| }, |
| "teacher_isolation": { |
| "prediction_input_binding_verified": true, |
| "prediction_lane": "camera-only", |
| "process_environment_audited": false, |
| "scope": "The evaluator verifies prediction documents, exact input hashes, and the separation of public inputs from the teacher root. This establishes an interface boundary only; it does not audit the process or host that produced the predictions.", |
| "teacher_access_declaration": "none-declared", |
| "teacher_root_separate_from_public_bundle": true |
| } |
| } |
|
|
