cy0307 commited on 18 days ago

Commit

3a10443

verified ·

1 Parent(s): eeac43c

Publish Ropedia Xperience-10M task baseline cards

Browse files

Files changed (49) hide show

PROJECT_STATUS.md +6 -3
RESEARCH_ROADMAP.md +10 -0
data/mirror_parity.json +620 -61
data/omni_finetune_verified_result.json +1 -1
data/omni_model_comparison.json +61 -6
data/project_packet.json +1 -1
data/project_status.json +15 -3
data/research_roadmap.json +1 -1
data/research_roadmap_interactive.json +1 -1
data/website_integrity.json +11 -11
docs/assets/charts/episode_task_scores.svg +12 -12
docs/assets/charts/episode_task_scores_minimal_vs_neural.svg +24 -24
docs/assets/charts/episode_task_scores_neural_mlp.svg +12 -12
docs/assets/charts/research_direction_coverage.svg +4 -4
docs/assets/task_architectures.png +2 -2
docs/assets/task_architectures.svg +12 -12
docs/assets/task_suite_infographic.png +2 -2
docs/data/mirror_parity.json +620 -61
docs/data/omni_finetune_verified_result.json +1 -1
docs/data/omni_model_comparison.json +61 -6
docs/data/project_packet.json +1 -1
docs/data/project_status.json +15 -3
docs/data/research_roadmap.json +1 -1
docs/data/research_roadmap_interactive.json +1 -1
docs/data/website_integrity.json +11 -11
docs/index.html +6 -6
metrics/mirror_parity.json +95 -95
metrics/omni_finetune_verified_result.json +1 -1
metrics/omni_model_comparison.json +61 -6
metrics/project_packet.json +1 -1
metrics/project_status.json +15 -3
metrics/research_roadmap.json +1 -1
metrics/research_roadmap_interactive.json +1 -1
metrics/website_integrity.json +11 -11
results/omni_finetune/OMNI_MODEL_COMPARISON.md +7 -5
results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/RUN_REPORT.md +19 -0
results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json +136 -0
results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/progress.jsonl +3 -0
results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/training_metadata.json +8 -0
results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/RUN_REPORT.md +35 -0
results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/progress.jsonl +3 -0
results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/training_contract_audit.json +78 -0
results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/training_metadata.json +47 -0
scripts/omni/audit_cosmos3_super_training_contract.py +406 -0
scripts/omni/build_omni_model_comparison.py +106 -9
scripts/omni/export_cosmos3_camera_pose_targets.py +250 -0
scripts/omni/pack_cosmos3_super_action_batch.py +459 -0
scripts/omni/run_qwen3_omni_v4_4epoch_8gpu.sh +105 -0
scripts/verify_live_publication.py +2 -2

PROJECT_STATUS.md CHANGED Viewed

@@ -22,7 +22,8 @@ scale-up readiness; it is not presented as final full-dataset model quality.
 | Audio contribution study | Verified | `scripts/audio_ablation_and_raw_upgrade.py`, `results/audio_ablation/`, `docs/data/audio_ablation_summary.json` | Audio variants are compared across all 12 task contracts; audio improves the primary metric on 6 of 12 tasks, and a 588-d audio-window representation improves over the baseline audio variant on 6 of 12 tasks. |
 | Research takeaways | Verified | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes. |
 | Research roadmap | Current | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | The roadmap connects public-sample task development to the final verified Qwen3-Omni diagnostic result, same-split baseline alignment, action/subtask error analysis, robustness runs, world/policy branches, and the future Xperience-native pretraining goal. |
-| Foundation-model plan | Current | `FOUNDATION_MODEL_PLAN.md`, `docs/data/foundation_model_plan.json` | Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit. |
 | Omni model extension contract | Current | `OMNI_MODEL_EXTENSION_CONTRACT.md`, `configs/omni_backbones/`, `scripts/omni/backbone_registry.py`, `scripts/omni/smoke_test_backbone_packaging.py` | Future model branches must keep the same episode split discipline, held-out metrics, validation gate, public-safe package contract, and explicit forbidden-artifact policy before reporting results. |
 | Xperience Embodied Foundation Model | Future goal | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` | A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model. |
 | Evaluation protocol | Verified | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts. |
@@ -83,8 +84,10 @@ scale-up readiness; it is not presented as final full-dataset model quality.
 - Audio contribution is evaluated across all 12 task contracts in
   `results/audio_ablation/`.
 - Foundation-model selection is now explicit: Qwen3-Omni is the immediate
-  trainable pilot, Cosmos 3 is the first world-model branch, and policy models
-  such as OpenVLA/openpi/GR00T wait for action-target conversion.
 - Future model branches should be added through the backbone registry and
   verified package contract, not by creating one-off result folders with
   incompatible metrics or publication rules.

 | Audio contribution study | Verified | `scripts/audio_ablation_and_raw_upgrade.py`, `results/audio_ablation/`, `docs/data/audio_ablation_summary.json` | Audio variants are compared across all 12 task contracts; audio improves the primary metric on 6 of 12 tasks, and a 588-d audio-window representation improves over the baseline audio variant on 6 of 12 tasks. |
 | Research takeaways | Verified | `RESEARCH_TAKEAWAYS.md`, `docs/data/research_takeaways.json`, `scripts/build_research_takeaways.py` | The main result interpretation is generated from committed metrics: chronological class shift, neural gains on dynamics/order/alignment, open retrieval/reconstruction problems, and the need for held-out episodes. |
 | Research roadmap | Current | `RESEARCH_ROADMAP.md`, `docs/data/research_roadmap.json` | The roadmap connects public-sample task development to the final verified Qwen3-Omni diagnostic result, same-split baseline alignment, action/subtask error analysis, robustness runs, world/policy branches, and the future Xperience-native pretraining goal. |
+| Foundation-model plan | Current | `FOUNDATION_MODEL_PLAN.md`, `docs/data/foundation_model_plan.json` | Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is added as the first world-model/action-generation branch; Cosmos3-Super now has camera-pose proxy action targets that pass the contract audit and a schema-only batch-packer smoke. The current target mode is forward-dynamics, so it supports vision-velocity training under action conditioning, not supervised action-token prediction. OpenVLA/openpi/GR00T are policy candidates after robot-compatible action targets are explicit. |
+| Cosmos3-Super action-target contract | Ready for forward-dynamics trainer implementation | `scripts/omni/export_cosmos3_camera_pose_targets.py`, `scripts/omni/pack_cosmos3_super_action_batch.py`, `results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/target_manifest.json`, `results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json`, `results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json` | The selected 128-episode JSONL is augmented with 3,808/3,808 valid `camera_pose` proxy `cosmos_action_target` records from SLAM pose deltas. The schema-only packer smoke confirms the current `forward_dynamics` target should supervise noisy vision tokens under camera-pose conditioning; it does not supervise `preds_action`. Remaining work is a pipeline-loaded packer check, one-sample forward-dynamics overfit, and a separate policy/inverse target export before claiming action-token prediction. |
 | Omni model extension contract | Current | `OMNI_MODEL_EXTENSION_CONTRACT.md`, `configs/omni_backbones/`, `scripts/omni/backbone_registry.py`, `scripts/omni/smoke_test_backbone_packaging.py` | Future model branches must keep the same episode split discipline, held-out metrics, validation gate, public-safe package contract, and explicit forbidden-artifact policy before reporting results. |
 | Xperience Embodied Foundation Model | Future goal | `XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md` | A future full-corpus pretraining plan describes target modules, objectives, staged scale-up, hardware ranges, and evaluation for a domain-specific embodied foundation model. |
 | Evaluation protocol | Verified | `EVALUATION_PROTOCOL.md`, `docs/data/evaluation_protocol.json`, `scripts/build_evaluation_protocol.py` | Windowing, chronological split, per-task metrics, leakage controls, and current limitations are generated from committed metric artifacts. |
 - Audio contribution is evaluated across all 12 task contracts in
   `results/audio_ablation/`.
 - Foundation-model selection is now explicit: Qwen3-Omni is the immediate
+  trainable pilot, Cosmos 3 is the first world-model branch, and Cosmos3-Super
+  has a camera-pose proxy forward-dynamics contract ready for trainer
+  implementation; policy models such as OpenVLA/openpi/GR00T still wait for
+  robot-compatible action-target conversion.
 - Future model branches should be added through the backbone registry and
   verified package contract, not by creating one-off result folders with
   incompatible metrics or publication rules.

RESEARCH_ROADMAP.md CHANGED Viewed

@@ -145,6 +145,16 @@ objectives: audio-visible alignment, future-window prediction,
 action-conditioned world modeling, synthetic-data usefulness tests, policy-style
 next action, contact, object relevance, and affordance reasoning.
 ### 7. Xperience Embodied Foundation Model Pretraining
 This stage is the long-term full-corpus goal. Instead of adapting an existing

 action-conditioned world modeling, synthetic-data usefulness tests, policy-style
 next action, contact, object relevance, and affordance reasoning.
+Current Cosmos3-Super status: a camera-pose proxy action target export now
+augments all 3,808 selected 128-episode windows and passes the contract audit.
+A schema-only batch-packer smoke confirms the current `forward_dynamics` target
+uses camera-pose actions as conditioning and should supervise noisy vision
+tokens, not `preds_action`. This is a trainer-readiness artifact, not a
+fine-tuned Cosmos weight release. The next Cosmos step is a pipeline-loaded
+packer check and one-sample forward-dynamics overfit before any 96/16/16 Super
+LoRA run; supervised action-token prediction needs a separate policy or
+inverse-dynamics target export.
 ### 7. Xperience Embodied Foundation Model Pretraining
 This stage is the long-term full-corpus goal. Instead of adapting an existing

data/mirror_parity.json CHANGED Viewed

@@ -1,16 +1,21 @@
 {
-  "status": "pass",
-  "generated_at_utc": "2026-06-07T15:49:31+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
-    "failure_count": 0,
-    "failures_by_surface": {}
   },
   "checks": [
     {
       "name": "repo_hf_space_artifact_model_data_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_visual_asset_parity",
@@ -18,19 +23,19 @@
     },
     {
       "name": "repo_hf_validator_script_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_website_html_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_diagnostic_result_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_quality_doc_parity",
-      "status": "pass"
     }
   ],
   "groups": [
@@ -346,12 +351,12 @@
     },
     {
       "name": "data/omni_finetune_verified_result.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
-        "bytes": 3628,
-        "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
       },
       "mirrors": {
         "hf_space": {
@@ -373,16 +378,38 @@
           "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         }
       },
-      "failures": []
     },
     {
       "name": "data/omni_model_comparison.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
-        "bytes": 48296,
-        "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
       },
       "mirrors": {
         "hf_space": {
@@ -404,7 +431,29 @@
           "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         }
       },
-      "failures": []
     },
     {
       "name": "data/project_brief.json",
@@ -470,12 +519,12 @@
     },
     {
       "name": "data/project_packet.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
-        "bytes": 8005,
-        "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
       },
       "mirrors": {
         "hf_space": {
@@ -497,16 +546,38 @@
           "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         }
       },
-      "failures": []
     },
     {
       "name": "data/project_status.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 16455,
-        "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
       },
       "mirrors": {
         "hf_space": {
@@ -528,7 +599,29 @@
           "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         }
       },
-      "failures": []
     },
     {
       "name": "data/publication_audit.json",
@@ -687,12 +780,12 @@
     },
     {
       "name": "data/research_roadmap.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 10133,
-        "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
       },
       "mirrors": {
         "hf_space": {
@@ -714,16 +807,38 @@
           "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         }
       },
-      "failures": []
     },
     {
       "name": "data/research_roadmap_interactive.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 143560,
-        "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
       },
       "mirrors": {
         "hf_space": {
@@ -745,7 +860,29 @@
           "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         }
       },
-      "failures": []
     },
     {
       "name": "data/research_takeaways.json",
@@ -1028,12 +1165,12 @@
     },
     {
       "name": "data/website_integrity.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
-        "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
       },
       "mirrors": {
         "hf_space": {
@@ -1055,7 +1192,29 @@
           "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         }
       },
-      "failures": []
     },
     {
       "name": "data/xperience10m_dataset_card_alignment.json",
@@ -1781,12 +1940,12 @@
     },
     {
       "name": "scripts/omni/build_omni_model_comparison.py",
-      "status": "pass",
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
-        "bytes": 30236,
-        "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
       },
       "mirrors": {
         "hf_artifacts": {
@@ -1802,7 +1961,22 @@
           "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         }
       },
-      "failures": []
     },
     {
       "name": "scripts/omni/prepare_qwen3_lora_hf_package.py",
@@ -2156,12 +2330,12 @@
     },
     {
       "name": "scripts/verify_live_publication.py",
-      "status": "pass",
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
-        "bytes": 36201,
-        "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
       },
       "mirrors": {
         "hf_artifacts": {
@@ -2177,7 +2351,22 @@
           "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         }
       },
-      "failures": []
     },
     {
       "name": "scripts/validate_mirror_parity.py",
@@ -2406,12 +2595,12 @@
     },
     {
       "name": "website/index.html",
-      "status": "pass",
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 180727,
-        "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
       },
       "mirrors": {
         "hf_space": {
@@ -2427,7 +2616,22 @@
           "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         }
       },
-      "failures": []
     },
     {
       "name": "website/research_roadmap.html",
@@ -2692,12 +2896,12 @@
     },
     {
       "name": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
-      "status": "pass",
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
-        "bytes": 9231,
-        "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
       },
       "mirrors": {
         "hf_space": {
@@ -2719,7 +2923,29 @@
           "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         }
       },
-      "failures": []
     },
     {
       "name": "results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md",
@@ -7032,12 +7258,12 @@
     },
     {
       "name": "docs/RESEARCH_ROADMAP.md",
-      "status": "pass",
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 12233,
-        "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
       },
       "mirrors": {
         "hf_space": {
@@ -7059,16 +7285,38 @@
           "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         }
       },
-      "failures": []
     },
     {
       "name": "docs/PROJECT_STATUS.md",
-      "status": "pass",
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 9926,
-        "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
       },
       "mirrors": {
         "hf_space": {
@@ -7090,7 +7338,29 @@
           "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         }
       },
-      "failures": []
     },
     {
       "name": "docs/PUBLIC_SURFACE_QA.md",
@@ -7217,5 +7487,294 @@
       "failures": []
     }
   ],
-  "failures": []
 }

 {
+  "status": "fail",
+  "generated_at_utc": "2026-06-07T17:27:20+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
+    "failure_count": 36,
+    "failures_by_surface": {
+      "hf_space": 11,
+      "hf_artifacts": 12,
+      "hf_model": 12,
+      "hf_artifacts_docs": 1
+    }
   },
   "checks": [
     {
       "name": "repo_hf_space_artifact_model_data_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_visual_asset_parity",
     },
     {
       "name": "repo_hf_validator_script_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_website_html_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_diagnostic_result_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_quality_doc_parity",
+      "status": "fail"
     }
   ],
   "groups": [
     },
     {
       "name": "data/omni_finetune_verified_result.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
+        "bytes": 3768,
+        "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        }
+      ]
     },
     {
       "name": "data/omni_model_comparison.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
+        "bytes": 50422,
+        "sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        }
+      ]
     },
     {
       "name": "data/project_brief.json",
     },
     {
       "name": "data/project_packet.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
+        "bytes": 8098,
+        "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        }
+      ]
     },
     {
       "name": "data/project_status.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
+        "bytes": 18062,
+        "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        }
+      ]
     },
     {
       "name": "data/publication_audit.json",
     },
     {
       "name": "data/research_roadmap.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
+        "bytes": 10246,
+        "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        }
+      ]
     },
     {
       "name": "data/research_roadmap_interactive.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
+        "bytes": 143673,
+        "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        }
+      ]
     },
     {
       "name": "data/research_takeaways.json",
     },
     {
       "name": "data/website_integrity.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
+        "sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        }
+      ]
     },
     {
       "name": "data/xperience10m_dataset_card_alignment.json",
     },
     {
       "name": "scripts/omni/build_omni_model_comparison.py",
+      "status": "fail",
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
+        "bytes": 35566,
+        "sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a"
       },
       "mirrors": {
         "hf_artifacts": {
           "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
+          "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+          "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
+          "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+          "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+        }
+      ]
     },
     {
       "name": "scripts/omni/prepare_qwen3_lora_hf_package.py",
     },
     {
       "name": "scripts/verify_live_publication.py",
+      "status": "fail",
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
+        "bytes": 36285,
+        "sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471"
       },
       "mirrors": {
         "hf_artifacts": {
           "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:scripts/verify_live_publication.py",
+          "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+          "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:scripts/verify_live_publication.py",
+          "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+          "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+        }
+      ]
     },
     {
       "name": "scripts/validate_mirror_parity.py",
     },
     {
       "name": "website/index.html",
+      "status": "fail",
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
+        "bytes": 181095,
+        "sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:index.html",
+          "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+          "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+        },
+        {
+          "surface": "hf_artifacts_docs",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/index.html",
+          "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+          "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+        }
+      ]
     },
     {
       "name": "website/research_roadmap.html",
     },
     {
       "name": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "status": "fail",
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
+        "bytes": 9893,
+        "sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        }
+      ]
     },
     {
       "name": "results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md",
     },
     {
       "name": "docs/RESEARCH_ROADMAP.md",
+      "status": "fail",
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
+        "bytes": 12874,
+        "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        }
+      ]
     },
     {
       "name": "docs/PROJECT_STATUS.md",
+      "status": "fail",
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
+        "bytes": 11369,
+        "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        }
+      ]
     },
     {
       "name": "docs/PUBLIC_SURFACE_QA.md",
       "failures": []
     }
   ],
+  "failures": [
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "scripts/omni/build_omni_model_comparison.py",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
+      "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+      "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+    },
+    {
+      "group": "scripts/omni/build_omni_model_comparison.py",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
+      "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+      "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+    },
+    {
+      "group": "scripts/verify_live_publication.py",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:scripts/verify_live_publication.py",
+      "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+      "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+    },
+    {
+      "group": "scripts/verify_live_publication.py",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:scripts/verify_live_publication.py",
+      "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+      "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+    },
+    {
+      "group": "website/index.html",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:index.html",
+      "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+      "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+    },
+    {
+      "group": "website/index.html",
+      "surface": "hf_artifacts_docs",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/index.html",
+      "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+      "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    }
+  ]
 }

data/omni_finetune_verified_result.json CHANGED Viewed

@@ -80,7 +80,7 @@
   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
-    "Implement the Cosmos3-Super diffusion/action target packer and supervised loss before claiming Cosmos3 fine-tuning.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
+    "Implement the Cosmos3-Super pipeline-loaded batch packer and one-sample forward-dynamics overfit before claiming Cosmos3 fine-tuning; camera-pose proxy targets are now exported, contract-audited, and schema-packed, but no Cosmos weights have been updated.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

data/omni_model_comparison.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
-  "generated_at_utc": "2026-06-07T15:34:51+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
@@ -8,7 +8,7 @@
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
-    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation rather than a new fine-tuned weight release."
   ],
   "versions": [
     {
@@ -1012,7 +1012,62 @@
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
-          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible, but blocks true fine-tuning until a Cosmos-specific diffusion/action target packer and supervised loss are implemented."
         }
       ],
       "multi_episode_128_runs": [
@@ -1056,7 +1111,7 @@
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
-      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. The readiness probe records why true Cosmos3-Super fine-tuning is not launched yet."
     }
   ],
   "model_group_reading_notes": [
@@ -1064,10 +1119,10 @@
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
-    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a training-readiness probe; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
-    "Promote Cosmos3 from Nano compatibility and Super base-weight evaluation to true fine-tuning only after a dedicated Cosmos diffusion/action target packer and supervised loss produce new weights."
   ]
 }

 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
+  "generated_at_utc": "2026-06-07T17:27:36+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
+    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation; Cosmos3-Super now has a camera-pose forward-dynamics contract audit and schema-only packer smoke, but no new fine-tuned weight release."
   ],
   "versions": [
     {
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
+          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible. It predates the camera-pose action-target export, so use the 20260608 contract audit for the current trainer-readiness status."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608",
+          "title": "Cosmos3-Super Camera-Pose Target Audit",
+          "scope_label": "action target contract",
+          "scope": "selected 128-episode 96/16/16 dataset augmented with camera_pose proxy cosmos_action_target records",
+          "status": "ready_for_forward_dynamics_trainer",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+          "split": "train/val/test by selected episode/session",
+          "counts": {
+            "dataset_samples": 3808,
+            "rows_with_action_target": 3808,
+            "valid_action_targets": 3808,
+            "split_counts": {
+              "train": 2848,
+              "val": 512,
+              "test": 448
+            },
+            "episode_split_counts": {
+              "test": 14,
+              "train": 89,
+              "val": 16
+            }
+          },
+          "primary_metrics": {
+            "domain_name": "camera_pose",
+            "raw_action_dim": 9,
+            "mode": "forward_dynamics",
+            "valid_action_targets": 3808,
+            "weights_updated": false
+          },
+          "weights": "none; action-target contract audit only, no adapter checkpoint",
+          "interpretation": "The selected dataset now has valid Cosmos3 camera_pose forward_dynamics targets for an egocentric camera-motion proxy. These remove the target-schema blocker for action-conditioned world-model training, but they supervise noisy vision tokens rather than preds_action. The remaining work is a pipeline-loaded packer check and one-sample forward-dynamics overfit; action-token prediction needs a separate policy or inverse-dynamics target export."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608",
+          "title": "Cosmos3-Super Action Batch Packer Smoke",
+          "scope_label": "batch packer",
+          "scope": "one selected train row from the camera_pose forward_dynamics augmented JSONL",
+          "status": "pass",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json",
+          "split": "train",
+          "counts": {
+            "samples": 1,
+            "raw_action_rows": 8,
+            "raw_action_dim": 9
+          },
+          "primary_metrics": {
+            "mode": "forward_dynamics",
+            "loss_surface": "vision_velocity_conditioned_on_camera_pose",
+            "pipeline_loaded": false,
+            "weights_updated": false
+          },
+          "weights": "none; schema-only packer smoke, no adapter checkpoint",
+          "interpretation": "The selected row maps to a camera_pose forward_dynamics contract. In the installed Cosmos3 pipeline this uses raw actions as conditioning and supervises noisy vision tokens; it does not supervise preds_action."
         }
       ],
       "multi_episode_128_runs": [
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
+      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. A camera-pose proxy forward-dynamics target export now passes the contract audit and schema-only packer smoke; true Cosmos3-Super fine-tuning is still not launched until the pipeline-loaded packer check and one-sample overfit exist."
     }
   ],
   "model_group_reading_notes": [
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
+    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a camera-pose forward-dynamics contract audit; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
+    "Promote Cosmos3 from Nano compatibility, Super base-weight evaluation, and the camera-pose forward-dynamics contract to true fine-tuning only after the pipeline-loaded packer check and one-sample overfit produce new weights."
   ]
 }

data/project_packet.json CHANGED Viewed

@@ -41,7 +41,7 @@
         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
-      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, and the Cosmos3-Super base-weight Reasoner evaluation are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
+      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, the Cosmos3-Super base-weight Reasoner evaluation, and the Cosmos3-Super camera-pose forward-dynamics contract audit plus schema-only packer smoke are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

data/project_status.json CHANGED Viewed

@@ -119,7 +119,7 @@
         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
-      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package plus a verified Cosmos3-Super base-weight Reasoner evaluation; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
@@ -244,6 +244,18 @@
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
@@ -276,11 +288,11 @@
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
-    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, and Cosmos3-Super Reasoner is verified as a base-weight evaluation; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
+      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package, a verified Cosmos3-Super base-weight Reasoner evaluation, and a Cosmos3-Super camera-pose proxy forward-dynamics contract audit plus schema-only packer smoke. The current target supports vision-velocity training under action conditioning, not supervised action-token prediction; OpenVLA/openpi/GR00T are policy candidates after robot-compatible action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
+    {
+      "area": "Cosmos3-Super action-target contract",
+      "status": "ready_for_forward_dynamics_trainer_implementation",
+      "evidence": [
+        "scripts/omni/export_cosmos3_camera_pose_targets.py",
+        "scripts/omni/pack_cosmos3_super_action_batch.py",
+        "results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/target_manifest.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json"
+      ],
+      "readout": "The selected 128-episode JSONL is augmented with 3,808/3,808 valid camera_pose proxy cosmos_action_target records from SLAM pose deltas. The schema-only packer smoke confirms the current forward_dynamics target should supervise noisy vision tokens under camera-pose conditioning; it does not supervise preds_action. Remaining work is a pipeline-loaded packer check, one-sample forward-dynamics overfit, and a separate policy/inverse target export before claiming action-token prediction."
+    },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
+    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, Cosmos3-Super Reasoner is verified as a base-weight evaluation, and Cosmos3-Super camera-pose forward-dynamics targets now pass the contract audit plus a schema-only packer smoke; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, Cosmos3-Super has a camera-pose proxy forward-dynamics contract ready for trainer implementation, and policy models such as OpenVLA/openpi/GR00T wait for robot-compatible action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

data/research_roadmap.json CHANGED Viewed

@@ -133,7 +133,7 @@
         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

data/research_roadmap_interactive.json CHANGED Viewed

@@ -2369,7 +2369,7 @@
       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets.",
       "stage": "omni",
       "status": "next"
     },

       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets.",
       "stage": "omni",
       "status": "next"
     },

data/website_integrity.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-07T15:47:32+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -75,7 +75,7 @@
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
-      "evidence_index": 90477
     },
     {
       "name": "project_status_links_json",
@@ -153,8 +153,8 @@
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
-      "protocol_index": 87160,
-      "evidence_index": 90477
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -292,7 +292,7 @@
     },
     {
       "path": "data/mirror_parity.json",
-      "bytes": 410374,
       "top_level_type": "dict"
     },
     {
@@ -302,12 +302,12 @@
     },
     {
       "path": "data/omni_finetune_verified_result.json",
-      "bytes": 3628,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
-      "bytes": 48296,
       "top_level_type": "dict"
     },
     {
@@ -322,12 +322,12 @@
     },
     {
       "path": "data/project_packet.json",
-      "bytes": 8005,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
-      "bytes": 16455,
       "top_level_type": "dict"
     },
     {
@@ -367,12 +367,12 @@
     },
     {
       "path": "data/research_roadmap.json",
-      "bytes": 10133,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
-      "bytes": 143560,
       "top_level_type": "dict"
     },
     {

 {
   "status": "pass",
+  "generated_at_utc": "2026-06-07T17:27:17+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
+      "evidence_index": 90659
     },
     {
       "name": "project_status_links_json",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
+      "protocol_index": 87218,
+      "evidence_index": 90659
     },
     {
       "name": "evaluation_protocol_links_json",
     },
     {
       "path": "data/mirror_parity.json",
+      "bytes": 319291,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/omni_finetune_verified_result.json",
+      "bytes": 3768,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
+      "bytes": 50422,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/project_packet.json",
+      "bytes": 8098,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
+      "bytes": 18062,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/research_roadmap.json",
+      "bytes": 10246,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
+      "bytes": 143673,
       "top_level_type": "dict"
     },
     {

docs/assets/charts/episode_task_scores.svg CHANGED Viewed

docs/assets/charts/episode_task_scores_minimal_vs_neural.svg CHANGED Viewed

docs/assets/charts/episode_task_scores_neural_mlp.svg CHANGED Viewed

docs/assets/charts/research_direction_coverage.svg CHANGED Viewed

docs/assets/task_architectures.png CHANGED Viewed

Git LFS Details

SHA256: 076c2e463ddce473e9138ac6f3615152d59031d6be2aa5c3d9ae1ace3d3f6c83
Pointer size: 131 Bytes
Size of remote file: 762 kB

Git LFS Details

SHA256: f08b03bc21e194efe382347d74cf89cd6ac65dede51889971dbfc2fb9d1de3c2
Pointer size: 131 Bytes
Size of remote file: 774 kB

docs/assets/task_architectures.svg CHANGED Viewed

docs/assets/task_suite_infographic.png CHANGED Viewed

Git LFS Details

SHA256: 213d81f49d27e3f2560c79e29a017c017cbe38d8d605815bf3bc87834a1424ae
Pointer size: 132 Bytes
Size of remote file: 2.61 MB

Git LFS Details

SHA256: 1275e2adaef920ecde7c29dc62c8d79d4f13475a0c09bc3baa693f47cdec2e1f
Pointer size: 132 Bytes
Size of remote file: 1.59 MB

docs/data/mirror_parity.json CHANGED Viewed

@@ -1,16 +1,21 @@
 {
-  "status": "pass",
-  "generated_at_utc": "2026-06-07T15:49:31+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
-    "failure_count": 0,
-    "failures_by_surface": {}
   },
   "checks": [
     {
       "name": "repo_hf_space_artifact_model_data_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_visual_asset_parity",
@@ -18,19 +23,19 @@
     },
     {
       "name": "repo_hf_validator_script_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_website_html_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_diagnostic_result_parity",
-      "status": "pass"
     },
     {
       "name": "repo_hf_quality_doc_parity",
-      "status": "pass"
     }
   ],
   "groups": [
@@ -346,12 +351,12 @@
     },
     {
       "name": "data/omni_finetune_verified_result.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
-        "bytes": 3628,
-        "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
       },
       "mirrors": {
         "hf_space": {
@@ -373,16 +378,38 @@
           "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         }
       },
-      "failures": []
     },
     {
       "name": "data/omni_model_comparison.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
-        "bytes": 48296,
-        "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
       },
       "mirrors": {
         "hf_space": {
@@ -404,7 +431,29 @@
           "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         }
       },
-      "failures": []
     },
     {
       "name": "data/project_brief.json",
@@ -470,12 +519,12 @@
     },
     {
       "name": "data/project_packet.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
-        "bytes": 8005,
-        "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
       },
       "mirrors": {
         "hf_space": {
@@ -497,16 +546,38 @@
           "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         }
       },
-      "failures": []
     },
     {
       "name": "data/project_status.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 16455,
-        "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
       },
       "mirrors": {
         "hf_space": {
@@ -528,7 +599,29 @@
           "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         }
       },
-      "failures": []
     },
     {
       "name": "data/publication_audit.json",
@@ -687,12 +780,12 @@
     },
     {
       "name": "data/research_roadmap.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 10133,
-        "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
       },
       "mirrors": {
         "hf_space": {
@@ -714,16 +807,38 @@
           "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         }
       },
-      "failures": []
     },
     {
       "name": "data/research_roadmap_interactive.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 143560,
-        "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
       },
       "mirrors": {
         "hf_space": {
@@ -745,7 +860,29 @@
           "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         }
       },
-      "failures": []
     },
     {
       "name": "data/research_takeaways.json",
@@ -1028,12 +1165,12 @@
     },
     {
       "name": "data/website_integrity.json",
-      "status": "pass",
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
-        "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
       },
       "mirrors": {
         "hf_space": {
@@ -1055,7 +1192,29 @@
           "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         }
       },
-      "failures": []
     },
     {
       "name": "data/xperience10m_dataset_card_alignment.json",
@@ -1781,12 +1940,12 @@
     },
     {
       "name": "scripts/omni/build_omni_model_comparison.py",
-      "status": "pass",
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
-        "bytes": 30236,
-        "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
       },
       "mirrors": {
         "hf_artifacts": {
@@ -1802,7 +1961,22 @@
           "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         }
       },
-      "failures": []
     },
     {
       "name": "scripts/omni/prepare_qwen3_lora_hf_package.py",
@@ -2156,12 +2330,12 @@
     },
     {
       "name": "scripts/verify_live_publication.py",
-      "status": "pass",
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
-        "bytes": 36201,
-        "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
       },
       "mirrors": {
         "hf_artifacts": {
@@ -2177,7 +2351,22 @@
           "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         }
       },
-      "failures": []
     },
     {
       "name": "scripts/validate_mirror_parity.py",
@@ -2406,12 +2595,12 @@
     },
     {
       "name": "website/index.html",
-      "status": "pass",
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 180727,
-        "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
       },
       "mirrors": {
         "hf_space": {
@@ -2427,7 +2616,22 @@
           "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         }
       },
-      "failures": []
     },
     {
       "name": "website/research_roadmap.html",
@@ -2692,12 +2896,12 @@
     },
     {
       "name": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
-      "status": "pass",
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
-        "bytes": 9231,
-        "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
       },
       "mirrors": {
         "hf_space": {
@@ -2719,7 +2923,29 @@
           "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         }
       },
-      "failures": []
     },
     {
       "name": "results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md",
@@ -7032,12 +7258,12 @@
     },
     {
       "name": "docs/RESEARCH_ROADMAP.md",
-      "status": "pass",
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 12233,
-        "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
       },
       "mirrors": {
         "hf_space": {
@@ -7059,16 +7285,38 @@
           "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         }
       },
-      "failures": []
     },
     {
       "name": "docs/PROJECT_STATUS.md",
-      "status": "pass",
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 9926,
-        "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
       },
       "mirrors": {
         "hf_space": {
@@ -7090,7 +7338,29 @@
           "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         }
       },
-      "failures": []
     },
     {
       "name": "docs/PUBLIC_SURFACE_QA.md",
@@ -7217,5 +7487,294 @@
       "failures": []
     }
   ],
-  "failures": []
 }

 {
+  "status": "fail",
+  "generated_at_utc": "2026-06-07T17:27:20+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
+    "failure_count": 36,
+    "failures_by_surface": {
+      "hf_space": 11,
+      "hf_artifacts": 12,
+      "hf_model": 12,
+      "hf_artifacts_docs": 1
+    }
   },
   "checks": [
     {
       "name": "repo_hf_space_artifact_model_data_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_visual_asset_parity",
     },
     {
       "name": "repo_hf_validator_script_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_website_html_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_diagnostic_result_parity",
+      "status": "fail"
     },
     {
       "name": "repo_hf_quality_doc_parity",
+      "status": "fail"
     }
   ],
   "groups": [
     },
     {
       "name": "data/omni_finetune_verified_result.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
+        "bytes": 3768,
+        "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/omni_finetune_verified_result.json",
+          "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+          "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+        }
+      ]
     },
     {
       "name": "data/omni_model_comparison.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
+        "bytes": 50422,
+        "sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/omni_model_comparison.json",
+          "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+          "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+        }
+      ]
     },
     {
       "name": "data/project_brief.json",
     },
     {
       "name": "data/project_packet.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
+        "bytes": 8098,
+        "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/project_packet.json",
+          "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+          "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+        }
+      ]
     },
     {
       "name": "data/project_status.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
+        "bytes": 18062,
+        "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/project_status.json",
+          "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+          "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+        }
+      ]
     },
     {
       "name": "data/publication_audit.json",
     },
     {
       "name": "data/research_roadmap.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
+        "bytes": 10246,
+        "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/research_roadmap.json",
+          "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+          "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+        }
+      ]
     },
     {
       "name": "data/research_roadmap_interactive.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
+        "bytes": 143673,
+        "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/research_roadmap_interactive.json",
+          "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+          "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+        }
+      ]
     },
     {
       "name": "data/research_takeaways.json",
     },
     {
       "name": "data/website_integrity.json",
+      "status": "fail",
       "local": {
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
+        "sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:data/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/data/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:metrics/website_integrity.json",
+          "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+          "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+        }
+      ]
     },
     {
       "name": "data/xperience10m_dataset_card_alignment.json",
     },
     {
       "name": "scripts/omni/build_omni_model_comparison.py",
+      "status": "fail",
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
+        "bytes": 35566,
+        "sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a"
       },
       "mirrors": {
         "hf_artifacts": {
           "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
+          "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+          "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
+          "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+          "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+        }
+      ]
     },
     {
       "name": "scripts/omni/prepare_qwen3_lora_hf_package.py",
     },
     {
       "name": "scripts/verify_live_publication.py",
+      "status": "fail",
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
+        "bytes": 36285,
+        "sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471"
       },
       "mirrors": {
         "hf_artifacts": {
           "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:scripts/verify_live_publication.py",
+          "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+          "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:scripts/verify_live_publication.py",
+          "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+          "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+        }
+      ]
     },
     {
       "name": "scripts/validate_mirror_parity.py",
     },
     {
       "name": "website/index.html",
+      "status": "fail",
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
+        "bytes": 181095,
+        "sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:index.html",
+          "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+          "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+        },
+        {
+          "surface": "hf_artifacts_docs",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:docs/index.html",
+          "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+          "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+        }
+      ]
     },
     {
       "name": "website/research_roadmap.html",
     },
     {
       "name": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "status": "fail",
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
+        "bytes": 9893,
+        "sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+          "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+          "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+        }
+      ]
     },
     {
       "name": "results/omni_finetune/multi_episode_128_task_baselines/BASELINE_ALIGNMENT_REPORT.md",
     },
     {
       "name": "docs/RESEARCH_ROADMAP.md",
+      "status": "fail",
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
+        "bytes": 12874,
+        "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:RESEARCH_ROADMAP.md",
+          "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+          "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+        }
+      ]
     },
     {
       "name": "docs/PROJECT_STATUS.md",
+      "status": "fail",
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
+        "bytes": 11369,
+        "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
       },
       "mirrors": {
         "hf_space": {
           "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         }
       },
+      "failures": [
+        {
+          "surface": "hf_space",
+          "kind": "hash_mismatch",
+          "path": "hf_space:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        },
+        {
+          "surface": "hf_artifacts",
+          "kind": "hash_mismatch",
+          "path": "hf_artifacts:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        },
+        {
+          "surface": "hf_model",
+          "kind": "hash_mismatch",
+          "path": "hf_model:PROJECT_STATUS.md",
+          "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+          "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+        }
+      ]
     },
     {
       "name": "docs/PUBLIC_SURFACE_QA.md",
       "failures": []
     }
   ],
+  "failures": [
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_finetune_verified_result.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/omni_finetune_verified_result.json",
+      "expected_sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1",
+      "actual_sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/omni_model_comparison.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/omni_model_comparison.json",
+      "expected_sha256": "71d32b81180c9acadcc614dff99256dcc6e560be08f1c6bd1a32487eed704ebb",
+      "actual_sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_packet.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/project_packet.json",
+      "expected_sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15",
+      "actual_sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/project_status.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/project_status.json",
+      "expected_sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8",
+      "actual_sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/research_roadmap.json",
+      "expected_sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06",
+      "actual_sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/research_roadmap_interactive.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/research_roadmap_interactive.json",
+      "expected_sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6",
+      "actual_sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:data/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/data/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "data/website_integrity.json",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:metrics/website_integrity.json",
+      "expected_sha256": "31d063e601db5ed64b8156f417f48d4e0474bc2b6c9088d875d3d0e18b6f4828",
+      "actual_sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
+    },
+    {
+      "group": "scripts/omni/build_omni_model_comparison.py",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
+      "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+      "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+    },
+    {
+      "group": "scripts/omni/build_omni_model_comparison.py",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
+      "expected_sha256": "c66d3d9dd32dd16203bb5a832d9bdafb985c44d3b4040cbd58cd08e77a70458a",
+      "actual_sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
+    },
+    {
+      "group": "scripts/verify_live_publication.py",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:scripts/verify_live_publication.py",
+      "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+      "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+    },
+    {
+      "group": "scripts/verify_live_publication.py",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:scripts/verify_live_publication.py",
+      "expected_sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471",
+      "actual_sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
+    },
+    {
+      "group": "website/index.html",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:index.html",
+      "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+      "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+    },
+    {
+      "group": "website/index.html",
+      "surface": "hf_artifacts_docs",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:docs/index.html",
+      "expected_sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1",
+      "actual_sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
+      "expected_sha256": "fa2129ff8775376674bb4550a6dac629baa9a48a0d49986f6bd33341c4a7bddb",
+      "actual_sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/RESEARCH_ROADMAP.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:RESEARCH_ROADMAP.md",
+      "expected_sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347",
+      "actual_sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_space",
+      "kind": "hash_mismatch",
+      "path": "hf_space:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_artifacts",
+      "kind": "hash_mismatch",
+      "path": "hf_artifacts:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    },
+    {
+      "group": "docs/PROJECT_STATUS.md",
+      "surface": "hf_model",
+      "kind": "hash_mismatch",
+      "path": "hf_model:PROJECT_STATUS.md",
+      "expected_sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114",
+      "actual_sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
+    }
+  ]
 }

docs/data/omni_finetune_verified_result.json CHANGED Viewed

@@ -80,7 +80,7 @@
   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
-    "Implement the Cosmos3-Super diffusion/action target packer and supervised loss before claiming Cosmos3 fine-tuning.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
+    "Implement the Cosmos3-Super pipeline-loaded batch packer and one-sample forward-dynamics overfit before claiming Cosmos3 fine-tuning; camera-pose proxy targets are now exported, contract-audited, and schema-packed, but no Cosmos weights have been updated.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

docs/data/omni_model_comparison.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
-  "generated_at_utc": "2026-06-07T15:34:51+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
@@ -8,7 +8,7 @@
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
-    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation rather than a new fine-tuned weight release."
   ],
   "versions": [
     {
@@ -1012,7 +1012,62 @@
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
-          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible, but blocks true fine-tuning until a Cosmos-specific diffusion/action target packer and supervised loss are implemented."
         }
       ],
       "multi_episode_128_runs": [
@@ -1056,7 +1111,7 @@
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
-      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. The readiness probe records why true Cosmos3-Super fine-tuning is not launched yet."
     }
   ],
   "model_group_reading_notes": [
@@ -1064,10 +1119,10 @@
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
-    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a training-readiness probe; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
-    "Promote Cosmos3 from Nano compatibility and Super base-weight evaluation to true fine-tuning only after a dedicated Cosmos diffusion/action target packer and supervised loss produce new weights."
   ]
 }

 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
+  "generated_at_utc": "2026-06-07T17:27:36+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
+    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation; Cosmos3-Super now has a camera-pose forward-dynamics contract audit and schema-only packer smoke, but no new fine-tuned weight release."
   ],
   "versions": [
     {
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
+          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible. It predates the camera-pose action-target export, so use the 20260608 contract audit for the current trainer-readiness status."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608",
+          "title": "Cosmos3-Super Camera-Pose Target Audit",
+          "scope_label": "action target contract",
+          "scope": "selected 128-episode 96/16/16 dataset augmented with camera_pose proxy cosmos_action_target records",
+          "status": "ready_for_forward_dynamics_trainer",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+          "split": "train/val/test by selected episode/session",
+          "counts": {
+            "dataset_samples": 3808,
+            "rows_with_action_target": 3808,
+            "valid_action_targets": 3808,
+            "split_counts": {
+              "train": 2848,
+              "val": 512,
+              "test": 448
+            },
+            "episode_split_counts": {
+              "test": 14,
+              "train": 89,
+              "val": 16
+            }
+          },
+          "primary_metrics": {
+            "domain_name": "camera_pose",
+            "raw_action_dim": 9,
+            "mode": "forward_dynamics",
+            "valid_action_targets": 3808,
+            "weights_updated": false
+          },
+          "weights": "none; action-target contract audit only, no adapter checkpoint",
+          "interpretation": "The selected dataset now has valid Cosmos3 camera_pose forward_dynamics targets for an egocentric camera-motion proxy. These remove the target-schema blocker for action-conditioned world-model training, but they supervise noisy vision tokens rather than preds_action. The remaining work is a pipeline-loaded packer check and one-sample forward-dynamics overfit; action-token prediction needs a separate policy or inverse-dynamics target export."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608",
+          "title": "Cosmos3-Super Action Batch Packer Smoke",
+          "scope_label": "batch packer",
+          "scope": "one selected train row from the camera_pose forward_dynamics augmented JSONL",
+          "status": "pass",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json",
+          "split": "train",
+          "counts": {
+            "samples": 1,
+            "raw_action_rows": 8,
+            "raw_action_dim": 9
+          },
+          "primary_metrics": {
+            "mode": "forward_dynamics",
+            "loss_surface": "vision_velocity_conditioned_on_camera_pose",
+            "pipeline_loaded": false,
+            "weights_updated": false
+          },
+          "weights": "none; schema-only packer smoke, no adapter checkpoint",
+          "interpretation": "The selected row maps to a camera_pose forward_dynamics contract. In the installed Cosmos3 pipeline this uses raw actions as conditioning and supervises noisy vision tokens; it does not supervise preds_action."
         }
       ],
       "multi_episode_128_runs": [
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
+      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. A camera-pose proxy forward-dynamics target export now passes the contract audit and schema-only packer smoke; true Cosmos3-Super fine-tuning is still not launched until the pipeline-loaded packer check and one-sample overfit exist."
     }
   ],
   "model_group_reading_notes": [
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
+    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a camera-pose forward-dynamics contract audit; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
+    "Promote Cosmos3 from Nano compatibility, Super base-weight evaluation, and the camera-pose forward-dynamics contract to true fine-tuning only after the pipeline-loaded packer check and one-sample overfit produce new weights."
   ]
 }

docs/data/project_packet.json CHANGED Viewed

@@ -41,7 +41,7 @@
         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
-      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, and the Cosmos3-Super base-weight Reasoner evaluation are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
+      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, the Cosmos3-Super base-weight Reasoner evaluation, and the Cosmos3-Super camera-pose forward-dynamics contract audit plus schema-only packer smoke are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

docs/data/project_status.json CHANGED Viewed

@@ -119,7 +119,7 @@
         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
-      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package plus a verified Cosmos3-Super base-weight Reasoner evaluation; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
@@ -244,6 +244,18 @@
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
@@ -276,11 +288,11 @@
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
-    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, and Cosmos3-Super Reasoner is verified as a base-weight evaluation; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
+      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package, a verified Cosmos3-Super base-weight Reasoner evaluation, and a Cosmos3-Super camera-pose proxy forward-dynamics contract audit plus schema-only packer smoke. The current target supports vision-velocity training under action conditioning, not supervised action-token prediction; OpenVLA/openpi/GR00T are policy candidates after robot-compatible action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
+    {
+      "area": "Cosmos3-Super action-target contract",
+      "status": "ready_for_forward_dynamics_trainer_implementation",
+      "evidence": [
+        "scripts/omni/export_cosmos3_camera_pose_targets.py",
+        "scripts/omni/pack_cosmos3_super_action_batch.py",
+        "results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/target_manifest.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json"
+      ],
+      "readout": "The selected 128-episode JSONL is augmented with 3,808/3,808 valid camera_pose proxy cosmos_action_target records from SLAM pose deltas. The schema-only packer smoke confirms the current forward_dynamics target should supervise noisy vision tokens under camera-pose conditioning; it does not supervise preds_action. Remaining work is a pipeline-loaded packer check, one-sample forward-dynamics overfit, and a separate policy/inverse target export before claiming action-token prediction."
+    },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
+    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, Cosmos3-Super Reasoner is verified as a base-weight evaluation, and Cosmos3-Super camera-pose forward-dynamics targets now pass the contract audit plus a schema-only packer smoke; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, Cosmos3-Super has a camera-pose proxy forward-dynamics contract ready for trainer implementation, and policy models such as OpenVLA/openpi/GR00T wait for robot-compatible action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

docs/data/research_roadmap.json CHANGED Viewed

@@ -133,7 +133,7 @@
         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

docs/data/research_roadmap_interactive.json CHANGED Viewed

@@ -2369,7 +2369,7 @@
       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets.",
       "stage": "omni",
       "status": "next"
     },

       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets.",
       "stage": "omni",
       "status": "next"
     },

docs/data/website_integrity.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-07T15:47:32+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -75,7 +75,7 @@
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
-      "evidence_index": 90477
     },
     {
       "name": "project_status_links_json",
@@ -153,8 +153,8 @@
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
-      "protocol_index": 87160,
-      "evidence_index": 90477
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -292,7 +292,7 @@
     },
     {
       "path": "data/mirror_parity.json",
-      "bytes": 410374,
       "top_level_type": "dict"
     },
     {
@@ -302,12 +302,12 @@
     },
     {
       "path": "data/omni_finetune_verified_result.json",
-      "bytes": 3628,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
-      "bytes": 48296,
       "top_level_type": "dict"
     },
     {
@@ -322,12 +322,12 @@
     },
     {
       "path": "data/project_packet.json",
-      "bytes": 8005,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
-      "bytes": 16455,
       "top_level_type": "dict"
     },
     {
@@ -367,12 +367,12 @@
     },
     {
       "path": "data/research_roadmap.json",
-      "bytes": 10133,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
-      "bytes": 143560,
       "top_level_type": "dict"
     },
     {

 {
   "status": "pass",
+  "generated_at_utc": "2026-06-07T17:27:17+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
+      "evidence_index": 90659
     },
     {
       "name": "project_status_links_json",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
+      "protocol_index": 87218,
+      "evidence_index": 90659
     },
     {
       "name": "evaluation_protocol_links_json",
     },
     {
       "path": "data/mirror_parity.json",
+      "bytes": 319291,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/omni_finetune_verified_result.json",
+      "bytes": 3768,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
+      "bytes": 50422,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/project_packet.json",
+      "bytes": 8098,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
+      "bytes": 18062,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/research_roadmap.json",
+      "bytes": 10246,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
+      "bytes": 143673,
       "top_level_type": "dict"
     },
     {

docs/index.html CHANGED Viewed

@@ -2409,7 +2409,7 @@
           <article class="roadmap-card" data-status="next">
             <span class="roadmap-status">next</span>
             <h3>Foundation-Model Selection Matrix</h3>
-            <p>Keep Qwen3-Omni as the first trainable held-out pilot, add Cosmos 3 for world modeling, and stage policy candidates after action targets are explicit.</p>
             <div class="roadmap-meta">
               <strong>Entry</strong><p>Completed 128-episode preparation or a smaller 3-8 episode preprocessing dry run.</p>
               <strong>Evidence</strong><p>Foundation model plan, source links, model-specific entry conditions, and evaluation additions.</p>
@@ -2488,8 +2488,8 @@
           <article class="artifact"><h3>Metric contract</h3><p>All 12 tasks list input, target, primary metric, minimal baseline score, and neural MLP score from committed result files.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Leakage controls</h3><p>Scalers fit on train windows only; future labels, target-side signals, caption/object labels, and contact labels stay on the target side unless explicitly queried.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/scripts/build_evaluation_protocol.py">builder script</a></article>
           <article class="artifact"><h3>Audio ablation</h3><p>Audio and no-audio variants are evaluated across all 12 task contracts under the same chronological split.</p><a href="data/audio_ablation_summary.json">audio summary</a></article>
-          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch, policy models wait for explicit action targets, and Xperience-native pretraining remains a later full-corpus goal.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
-          <article class="artifact"><h3>Next evaluation stage</h3><p>This public-sample run covers single-episode task development. The selected multi-episode Qwen3-Omni final diagnostic result is verified and meets the JSON-validity target; Cosmos3-Nano has a verified future-window compatibility package; and Cosmos3-Super has a verified base-weight Reasoner JSON-task evaluation. The next stage is action/subtask error analysis, true Cosmos fine-tuning, and policy-target conversion.</p><a href="data/omni_model_comparison.json">result comparison</a></article>
           <article class="artifact"><h3>Scale-up requirement</h3><p>Future Omni, Cosmos, and policy branches use the same episode split discipline, training metadata, held-out predictions, metrics, run report, and public-safe package gate.</p><a href="data/foundation_model_plan.json">scale-up status</a></article>
         </div>
       </div>
@@ -2542,7 +2542,7 @@
           <article class="evidence-card">
             <span class="status-pill">current plan</span>
             <h3>Foundation backbones are separated by role</h3>
-            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch; OpenVLA/openpi/GR00T are policy candidates after action-space conversion; Xperience-native pretraining is the later full-corpus goal.</p>
             <div class="evidence-links">
               <a href="data/foundation_model_plan.json">foundation model plan</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/FOUNDATION_MODEL_PLAN.md">plan doc</a>
@@ -2552,7 +2552,7 @@
           <article class="evidence-card">
             <span class="status-pill">verified diagnostic</span>
             <h3>Qwen3-Omni and Cosmos3 branches</h3>
-            <p>The selected 96/16/16 episode split produced verified Qwen3-Omni packages with 448 held-out test predictions. Cosmos3-Nano has 378 held-out future-window predictions, and Cosmos3-Super Reasoner has 448 held-out base-weight JSON-task predictions.</p>
             <div class="evidence-links">
               <a href="data/omni_model_comparison.json">result comparison</a>
               <a href="data/omni_finetune_verified_result.json">pilot result</a>
@@ -3160,7 +3160,7 @@
               <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style policy candidates, and the future Xperience-native pretraining goal.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
               <article class="artifact"><h3>Multi-episode data access</h3><p>Public data-access path, selected 128-episode pilot plan, and preparation requirements.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a></article>
               <article class="artifact"><h3>Qwen3-Omni LoRA group</h3><p>Separates the 1-episode sensor-adapter smoke test from the current 128-episode LoRA adapter package and older diagnostics.</p><a href="data/omni_model_comparison.json">Qwen group</a></article>
-              <article class="artifact"><h3>Cosmos3 groups</h3><p>Shows the verified Nano future-window compatibility package and the Super base-weight Reasoner JSON-task evaluation; neither is a new fine-tuned Cosmos weight release.</p><a href="data/omni_model_comparison.json">Cosmos groups</a></article>
               <article class="artifact"><h3>Scale-up requirement</h3><p>Future runs need validation tracking, held-out predictions, quality-target reporting, and the same public-safe package gate.</p><a href="data/foundation_model_plan.json">training requirements</a></article>
               <article class="artifact"><h3>Xperience-native pretraining</h3><p>Future plan for a domain-specific embodied foundation model trained from scratch over full-corpus video, audio, geometry, motion, inertial, and language streams.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
             </div>

           <article class="roadmap-card" data-status="next">
             <span class="roadmap-status">next</span>
             <h3>Foundation-Model Selection Matrix</h3>
+            <p>Keep Qwen3-Omni as the first trainable held-out pilot, use Cosmos 3 for world modeling and forward-dynamics trainer development, and stage policy candidates after robot-compatible action targets are explicit.</p>
             <div class="roadmap-meta">
               <strong>Entry</strong><p>Completed 128-episode preparation or a smaller 3-8 episode preprocessing dry run.</p>
               <strong>Evidence</strong><p>Foundation model plan, source links, model-specific entry conditions, and evaluation additions.</p>
           <article class="artifact"><h3>Metric contract</h3><p>All 12 tasks list input, target, primary metric, minimal baseline score, and neural MLP score from committed result files.</p><a href="data/summary_metrics.json">summary metrics</a></article>
           <article class="artifact"><h3>Leakage controls</h3><p>Scalers fit on train windows only; future labels, target-side signals, caption/object labels, and contact labels stay on the target side unless explicitly queried.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/scripts/build_evaluation_protocol.py">builder script</a></article>
           <article class="artifact"><h3>Audio ablation</h3><p>Audio and no-audio variants are evaluated across all 12 task contracts under the same chronological split.</p><a href="data/audio_ablation_summary.json">audio summary</a></article>
+          <article class="artifact"><h3>Foundation branch selection</h3><p>Qwen3-Omni is the first trainable baseline, Cosmos 3 becomes the world-model branch with a camera-pose proxy forward-dynamics contract ready for trainer work, policy models wait for robot-compatible action targets, and Xperience-native pretraining remains a later full-corpus goal.</p><a href="data/foundation_model_plan.json">backbone plan</a></article>
+          <article class="artifact"><h3>Next evaluation stage</h3><p>This public-sample run covers single-episode task development. The selected multi-episode Qwen3-Omni final diagnostic result is verified and meets the JSON-validity target; Cosmos3-Nano has a verified future-window compatibility package; and Cosmos3-Super has a verified base-weight JSON-task evaluation plus a camera-pose forward-dynamics contract audit. The next stage is action/subtask error analysis, true Cosmos fine-tuning, and policy-target conversion.</p><a href="data/omni_model_comparison.json">result comparison</a></article>
           <article class="artifact"><h3>Scale-up requirement</h3><p>Future Omni, Cosmos, and policy branches use the same episode split discipline, training metadata, held-out predictions, metrics, run report, and public-safe package gate.</p><a href="data/foundation_model_plan.json">scale-up status</a></article>
         </div>
       </div>
           <article class="evidence-card">
             <span class="status-pill">current plan</span>
             <h3>Foundation backbones are separated by role</h3>
+            <p>Qwen3-Omni stays first for held-out LoRA; Cosmos 3 is the world-model branch with camera-pose proxy forward-dynamics targets ready for trainer work; OpenVLA/openpi/GR00T are policy candidates after robot-compatible action conversion; Xperience-native pretraining is the later full-corpus goal.</p>
             <div class="evidence-links">
               <a href="data/foundation_model_plan.json">foundation model plan</a>
               <a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/FOUNDATION_MODEL_PLAN.md">plan doc</a>
           <article class="evidence-card">
             <span class="status-pill">verified diagnostic</span>
             <h3>Qwen3-Omni and Cosmos3 branches</h3>
+            <p>The selected 96/16/16 episode split produced verified Qwen3-Omni packages with 448 held-out test predictions. Cosmos3-Nano has 378 held-out future-window predictions, and Cosmos3-Super Reasoner has 448 held-out base-weight JSON-task predictions plus a camera-pose forward-dynamics contract audit.</p>
             <div class="evidence-links">
               <a href="data/omni_model_comparison.json">result comparison</a>
               <a href="data/omni_finetune_verified_result.json">pilot result</a>
               <article class="artifact"><h3>Foundation-model plan</h3><p>Backbone selection matrix covering Qwen3-Omni, Cosmos 3, GR00T, OpenVLA/openpi, Gemini Robotics, Octo, SmolVLA-style policy candidates, and the future Xperience-native pretraining goal.</p><a href="data/foundation_model_plan.json">foundation model plan</a></article>
               <article class="artifact"><h3>Multi-episode data access</h3><p>Public data-access path, selected 128-episode pilot plan, and preparation requirements.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/results/omni_finetune/MULTI_EPISODE_ACCESS_STATUS.md">data access</a></article>
               <article class="artifact"><h3>Qwen3-Omni LoRA group</h3><p>Separates the 1-episode sensor-adapter smoke test from the current 128-episode LoRA adapter package and older diagnostics.</p><a href="data/omni_model_comparison.json">Qwen group</a></article>
+              <article class="artifact"><h3>Cosmos3 groups</h3><p>Shows the verified Nano future-window compatibility package, the Super base-weight Reasoner JSON-task evaluation, and the Super camera-pose forward-dynamics contract audit; none is a new fine-tuned Cosmos weight release.</p><a href="data/omni_model_comparison.json">Cosmos groups</a></article>
               <article class="artifact"><h3>Scale-up requirement</h3><p>Future runs need validation tracking, held-out predictions, quality-target reporting, and the same public-safe package gate.</p><a href="data/foundation_model_plan.json">training requirements</a></article>
               <article class="artifact"><h3>Xperience-native pretraining</h3><p>Future plan for a domain-specific embodied foundation model trained from scratch over full-corpus video, audio, geometry, motion, inertial, and language streams.</p><a href="https://github.com/ChaoYue0307/ropedia-xperience-10m-task-suite/blob/main/XPERIENCE_EMBODIED_FOUNDATION_MODEL_PRETRAINING.md">pretraining plan</a></article>
             </div>

metrics/mirror_parity.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-07T15:49:31+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
@@ -350,27 +350,27 @@
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
-        "bytes": 3628,
-        "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/omni_finetune_verified_result.json",
           "exists": true,
-          "bytes": 3628,
-          "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
           "exists": true,
-          "bytes": 3628,
-          "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         },
         "hf_model": {
           "path": "hf_model:metrics/omni_finetune_verified_result.json",
           "exists": true,
-          "bytes": 3628,
-          "sha256": "ce28a11876aa33feb1f7b28c977c1d3e708b7d5d8b24b062684d472ba671d004"
         }
       },
       "failures": []
@@ -381,27 +381,27 @@
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
-        "bytes": 48296,
-        "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/omni_model_comparison.json",
           "exists": true,
-          "bytes": 48296,
-          "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/omni_model_comparison.json",
           "exists": true,
-          "bytes": 48296,
-          "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         },
         "hf_model": {
           "path": "hf_model:metrics/omni_model_comparison.json",
           "exists": true,
-          "bytes": 48296,
-          "sha256": "1c968bd58842af9a4e6159c1a8bd171aec08757bb77fce9f04c55030be08357f"
         }
       },
       "failures": []
@@ -474,27 +474,27 @@
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
-        "bytes": 8005,
-        "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_packet.json",
           "exists": true,
-          "bytes": 8005,
-          "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_packet.json",
           "exists": true,
-          "bytes": 8005,
-          "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_packet.json",
           "exists": true,
-          "bytes": 8005,
-          "sha256": "2258fecb80850c745e60cb28733869c49a5182879d9d0461b666a5575e3c1610"
         }
       },
       "failures": []
@@ -505,27 +505,27 @@
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
-        "bytes": 16455,
-        "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
-          "bytes": 16455,
-          "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
-          "bytes": 16455,
-          "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
-          "bytes": 16455,
-          "sha256": "3590ee1e09ecf819080a7714ea9629db305e1fd68c99a65f62bb65061c0d766c"
         }
       },
       "failures": []
@@ -691,27 +691,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
-        "bytes": 10133,
-        "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap.json",
           "exists": true,
-          "bytes": 10133,
-          "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap.json",
           "exists": true,
-          "bytes": 10133,
-          "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap.json",
           "exists": true,
-          "bytes": 10133,
-          "sha256": "45fd3a1bde93654ccfe14f9271928a67b36eb3f166826bfbdbb9c1092ad33bcf"
         }
       },
       "failures": []
@@ -722,27 +722,27 @@
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
-        "bytes": 143560,
-        "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 143560,
-          "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 143560,
-          "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap_interactive.json",
           "exists": true,
-          "bytes": 143560,
-          "sha256": "9198752056a40eb5a7457ded21576862d9954be1f0f4a9e996e935d328ef4062"
         }
       },
       "failures": []
@@ -1033,26 +1033,26 @@
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
-        "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
           "bytes": 15375,
-          "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 15375,
-          "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
           "bytes": 15375,
-          "sha256": "449b5525a0fc9ba200e59c6248e5ce963381938ab2c2027e1933db9483622037"
         }
       },
       "failures": []
@@ -1785,21 +1785,21 @@
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
-        "bytes": 30236,
-        "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
           "exists": true,
-          "bytes": 30236,
-          "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         },
         "hf_model": {
           "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
           "exists": true,
-          "bytes": 30236,
-          "sha256": "207b0bbfbea1cd3d7e6e77e7eafcf231b71c9f6483ffc36889234c7bafbcb1df"
         }
       },
       "failures": []
@@ -2160,21 +2160,21 @@
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
-        "bytes": 36201,
-        "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/verify_live_publication.py",
           "exists": true,
-          "bytes": 36201,
-          "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         },
         "hf_model": {
           "path": "hf_model:scripts/verify_live_publication.py",
           "exists": true,
-          "bytes": 36201,
-          "sha256": "76f03885867a8ed7095958a6948cbce81b4958fb74a09df24c24ad7eb5b0d944"
         }
       },
       "failures": []
@@ -2410,21 +2410,21 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 180727,
-        "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 180727,
-          "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 180727,
-          "sha256": "a88769e505d5af34674278f282ed1f482cc91dc711ddc0ed894a3fca5d08ff67"
         }
       },
       "failures": []
@@ -2696,27 +2696,27 @@
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
-        "bytes": 9231,
-        "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
-          "bytes": 9231,
-          "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
-          "bytes": 9231,
-          "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         },
         "hf_model": {
           "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
-          "bytes": 9231,
-          "sha256": "c38d12e138193f7200800d4dd8c149497de2c5f5895299e22fe81285b69fc62d"
         }
       },
       "failures": []
@@ -7036,27 +7036,27 @@
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
-        "bytes": 12233,
-        "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 12233,
-          "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 12233,
-          "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         },
         "hf_model": {
           "path": "hf_model:RESEARCH_ROADMAP.md",
           "exists": true,
-          "bytes": 12233,
-          "sha256": "020512aa647cef7d63eccf7bb8dd6cb86f0e5c457f3c0e3d5ef293e7b35a58bf"
         }
       },
       "failures": []
@@ -7067,27 +7067,27 @@
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
-        "bytes": 9926,
-        "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 9926,
-          "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 9926,
-          "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
-          "bytes": 9926,
-          "sha256": "c7dfb7a45f0c1ea435c16d93208a82da4227336e34f56a96d4afa04fce42438c"
         }
       },
       "failures": []

 {
   "status": "pass",
+  "generated_at_utc": "2026-06-07T17:31:58+00:00",
   "hf_root": "hf_publish",
   "summary": {
     "group_count": 234,
       "local": {
         "path": "repo:docs/data/omni_finetune_verified_result.json",
         "exists": true,
+        "bytes": 3768,
+        "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/omni_finetune_verified_result.json",
           "exists": true,
+          "bytes": 3768,
+          "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/omni_finetune_verified_result.json",
           "exists": true,
+          "bytes": 3768,
+          "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
         },
         "hf_model": {
           "path": "hf_model:metrics/omni_finetune_verified_result.json",
           "exists": true,
+          "bytes": 3768,
+          "sha256": "efc1b9c1938f358f44e2cfbc53bb395714217f8e158ecc0e2609a775c670c6e1"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/data/omni_model_comparison.json",
         "exists": true,
+        "bytes": 51589,
+        "sha256": "ba400d7c5dadd5fa654f3ba2b202be7f11537c1de7e2abee600ca431de2785a4"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/omni_model_comparison.json",
           "exists": true,
+          "bytes": 51589,
+          "sha256": "ba400d7c5dadd5fa654f3ba2b202be7f11537c1de7e2abee600ca431de2785a4"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/omni_model_comparison.json",
           "exists": true,
+          "bytes": 51589,
+          "sha256": "ba400d7c5dadd5fa654f3ba2b202be7f11537c1de7e2abee600ca431de2785a4"
         },
         "hf_model": {
           "path": "hf_model:metrics/omni_model_comparison.json",
           "exists": true,
+          "bytes": 51589,
+          "sha256": "ba400d7c5dadd5fa654f3ba2b202be7f11537c1de7e2abee600ca431de2785a4"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/data/project_packet.json",
         "exists": true,
+        "bytes": 8098,
+        "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_packet.json",
           "exists": true,
+          "bytes": 8098,
+          "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_packet.json",
           "exists": true,
+          "bytes": 8098,
+          "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_packet.json",
           "exists": true,
+          "bytes": 8098,
+          "sha256": "77cabac65b31db4e0477e20b1e6dfb06572bee42d8f71ac48f9380c0f4d86e15"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/data/project_status.json",
         "exists": true,
+        "bytes": 18062,
+        "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/project_status.json",
           "exists": true,
+          "bytes": 18062,
+          "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/project_status.json",
           "exists": true,
+          "bytes": 18062,
+          "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
         },
         "hf_model": {
           "path": "hf_model:metrics/project_status.json",
           "exists": true,
+          "bytes": 18062,
+          "sha256": "3f75b0894d215e39f69b4a477c06132eba00d4ed67cf6e39a22716e08ee725b8"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/data/research_roadmap.json",
         "exists": true,
+        "bytes": 10246,
+        "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap.json",
           "exists": true,
+          "bytes": 10246,
+          "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap.json",
           "exists": true,
+          "bytes": 10246,
+          "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap.json",
           "exists": true,
+          "bytes": 10246,
+          "sha256": "d34d763c3e880002f0b5de554b1b3f17b65f2cff24c5bc080ece938d04db2d06"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/data/research_roadmap_interactive.json",
         "exists": true,
+        "bytes": 143673,
+        "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/research_roadmap_interactive.json",
           "exists": true,
+          "bytes": 143673,
+          "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/research_roadmap_interactive.json",
           "exists": true,
+          "bytes": 143673,
+          "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
         },
         "hf_model": {
           "path": "hf_model:metrics/research_roadmap_interactive.json",
           "exists": true,
+          "bytes": 143673,
+          "sha256": "ad989e7cf78a213543614e23f90d4f03e5f5617b3ec6be43dfcc4b3a22cd6ac6"
         }
       },
       "failures": []
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 15375,
+        "sha256": "29b9ad18c3c76ebf8d453a77c726f2d56c207ea262d74a8b6d086092020bef94"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
           "bytes": 15375,
+          "sha256": "29b9ad18c3c76ebf8d453a77c726f2d56c207ea262d74a8b6d086092020bef94"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 15375,
+          "sha256": "29b9ad18c3c76ebf8d453a77c726f2d56c207ea262d74a8b6d086092020bef94"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
           "bytes": 15375,
+          "sha256": "29b9ad18c3c76ebf8d453a77c726f2d56c207ea262d74a8b6d086092020bef94"
         }
       },
       "failures": []
       "local": {
         "path": "repo:scripts/omni/build_omni_model_comparison.py",
         "exists": true,
+        "bytes": 35577,
+        "sha256": "593fa7179d2ad0ca03aa11652f3273f046468d38447a6f05b0c8f36c4be25889"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/omni/build_omni_model_comparison.py",
           "exists": true,
+          "bytes": 35577,
+          "sha256": "593fa7179d2ad0ca03aa11652f3273f046468d38447a6f05b0c8f36c4be25889"
         },
         "hf_model": {
           "path": "hf_model:scripts/omni/build_omni_model_comparison.py",
           "exists": true,
+          "bytes": 35577,
+          "sha256": "593fa7179d2ad0ca03aa11652f3273f046468d38447a6f05b0c8f36c4be25889"
         }
       },
       "failures": []
       "local": {
         "path": "repo:scripts/verify_live_publication.py",
         "exists": true,
+        "bytes": 36285,
+        "sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471"
       },
       "mirrors": {
         "hf_artifacts": {
           "path": "hf_artifacts:scripts/verify_live_publication.py",
           "exists": true,
+          "bytes": 36285,
+          "sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471"
         },
         "hf_model": {
           "path": "hf_model:scripts/verify_live_publication.py",
           "exists": true,
+          "bytes": 36285,
+          "sha256": "4605124056ca329069b1ec848372dda439258140e0e2aeb449d7bf1929623471"
         }
       },
       "failures": []
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
+        "bytes": 181095,
+        "sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
+          "bytes": 181095,
+          "sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
+          "bytes": 181095,
+          "sha256": "856d5f9529fc30adbd995f45df43af0861f5e48b8fbfb14cb4e4313ede097dc1"
         }
       },
       "failures": []
       "local": {
         "path": "repo:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
         "exists": true,
+        "bytes": 10215,
+        "sha256": "a5de891b2119941e27af8d28fd6d93c53387cc7609dea8fe4fe8e30786e1cc7c"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
+          "bytes": 10215,
+          "sha256": "a5de891b2119941e27af8d28fd6d93c53387cc7609dea8fe4fe8e30786e1cc7c"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
+          "bytes": 10215,
+          "sha256": "a5de891b2119941e27af8d28fd6d93c53387cc7609dea8fe4fe8e30786e1cc7c"
         },
         "hf_model": {
           "path": "hf_model:results/omni_finetune/OMNI_MODEL_COMPARISON.md",
           "exists": true,
+          "bytes": 10215,
+          "sha256": "a5de891b2119941e27af8d28fd6d93c53387cc7609dea8fe4fe8e30786e1cc7c"
         }
       },
       "failures": []
       "local": {
         "path": "repo:RESEARCH_ROADMAP.md",
         "exists": true,
+        "bytes": 12874,
+        "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:RESEARCH_ROADMAP.md",
           "exists": true,
+          "bytes": 12874,
+          "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:RESEARCH_ROADMAP.md",
           "exists": true,
+          "bytes": 12874,
+          "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
         },
         "hf_model": {
           "path": "hf_model:RESEARCH_ROADMAP.md",
           "exists": true,
+          "bytes": 12874,
+          "sha256": "834317a5b066b46046042be3f0c9ac7d12226a95728bd4a0a5898c3c96044347"
         }
       },
       "failures": []
       "local": {
         "path": "repo:PROJECT_STATUS.md",
         "exists": true,
+        "bytes": 11369,
+        "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:PROJECT_STATUS.md",
           "exists": true,
+          "bytes": 11369,
+          "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:PROJECT_STATUS.md",
           "exists": true,
+          "bytes": 11369,
+          "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
         },
         "hf_model": {
           "path": "hf_model:PROJECT_STATUS.md",
           "exists": true,
+          "bytes": 11369,
+          "sha256": "9ada29f7e7c8f6203abe2ddde67fcbe35656fa0c299b70d6adbd28053f69d114"
         }
       },
       "failures": []

metrics/omni_finetune_verified_result.json CHANGED Viewed

@@ -80,7 +80,7 @@
   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
-    "Implement the Cosmos3-Super diffusion/action target packer and supervised loss before claiming Cosmos3 fine-tuning.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

   "required_next_steps": [
     "Use the v3 strict-label predictions for action/subtask error analysis and unseen-label debugging.",
     "Keep the existing Qwen LoRA adapter repository as the weight-bearing artifact; v3 is an evaluation/package refresh over the same adapter, not new weights.",
+    "Implement the Cosmos3-Super pipeline-loaded batch packer and one-sample forward-dynamics overfit before claiming Cosmos3 fine-tuning; camera-pose proxy targets are now exported, contract-audited, and schema-packed, but no Cosmos weights have been updated.",
     "Use sharded Qwen eval for future long held-out passes to improve GPU utilization."
   ]
 }

metrics/omni_model_comparison.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
-  "generated_at_utc": "2026-06-07T15:34:51+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
@@ -8,7 +8,7 @@
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
-    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation rather than a new fine-tuned weight release."
   ],
   "versions": [
     {
@@ -1012,7 +1012,62 @@
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
-          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible, but blocks true fine-tuning until a Cosmos-specific diffusion/action target packer and supervised loss are implemented."
         }
       ],
       "multi_episode_128_runs": [
@@ -1056,7 +1111,7 @@
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
-      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. The readiness probe records why true Cosmos3-Super fine-tuning is not launched yet."
     }
   ],
   "model_group_reading_notes": [
@@ -1064,10 +1119,10 @@
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
-    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a training-readiness probe; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
-    "Promote Cosmos3 from Nano compatibility and Super base-weight evaluation to true fine-tuning only after a dedicated Cosmos diffusion/action target packer and supervised loss produce new weights."
   ]
 }

 {
   "title": "Ropedia Xperience-10M Current Result Versions and Model Groups",
+  "generated_at_utc": "2026-06-07T17:29:16+00:00",
   "status": "pass",
   "version_count": 3,
   "model_group_count": 4,
   "version_reading_notes": [
     "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
     "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
+    "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation; Cosmos3-Super now has a camera-pose forward-dynamics contract audit and schema-only packer smoke, but no new fine-tuned weight release."
   ],
   "versions": [
     {
             "weights_updated": false
           },
           "weights": "none; readiness audit only, no adapter checkpoint",
+          "interpretation": "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and the same JSON QA dataset are visible. It predates the camera-pose action-target export, so use the 20260608 contract audit for the current trainer-readiness status."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608",
+          "title": "Cosmos3-Super Camera-Pose Target Audit",
+          "scope_label": "action target contract",
+          "scope": "selected 128-episode 96/16/16 dataset augmented with camera_pose proxy cosmos_action_target records",
+          "status": "ready_for_forward_dynamics_trainer",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+          "split": "train/val/test by selected episode/session",
+          "counts": {
+            "dataset_samples": 3808,
+            "rows_with_action_target": 3808,
+            "valid_action_targets": 3808,
+            "split_counts": {
+              "train": 2848,
+              "val": 512,
+              "test": 448
+            },
+            "episode_split_counts": {
+              "test": 14,
+              "train": 89,
+              "val": 16
+            }
+          },
+          "primary_metrics": {
+            "domain_name": "camera_pose",
+            "raw_action_dim": 9,
+            "mode": "forward_dynamics",
+            "valid_action_targets": 3808,
+            "weights_updated": false
+          },
+          "weights": "none; action-target contract audit only, no adapter checkpoint",
+          "interpretation": "The selected dataset now has valid Cosmos3 camera_pose forward_dynamics targets for an egocentric camera-motion proxy. These remove the target-schema blocker for action-conditioned world-model training, but they supervise noisy vision tokens rather than preds_action. The remaining work is a pipeline-loaded packer check and one-sample forward-dynamics overfit; action-token prediction needs a separate policy or inverse-dynamics target export."
+        },
+        {
+          "id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608",
+          "title": "Cosmos3-Super Action Batch Packer Smoke",
+          "scope_label": "batch packer",
+          "scope": "one selected train row from the camera_pose forward_dynamics augmented JSONL",
+          "status": "pass",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json",
+          "split": "train",
+          "counts": {
+            "samples": 1,
+            "raw_action_rows": 8,
+            "raw_action_dim": 9
+          },
+          "primary_metrics": {
+            "mode": "forward_dynamics",
+            "loss_surface": "vision_velocity_conditioned_on_camera_pose",
+            "pipeline_loaded": false,
+            "weights_updated": false
+          },
+          "weights": "none; schema-only packer smoke, no adapter checkpoint",
+          "interpretation": "The selected row maps to a camera_pose forward_dynamics contract. In the installed Cosmos3 pipeline this uses raw actions as conditioning and supervises noisy vision tokens; it does not supervise preds_action."
         }
       ],
       "multi_episode_128_runs": [
           "weights_repository": "none for this run: staged base nv-community/Cosmos3-Super weights were evaluated through vLLM; create a separate repo only after new adapter or fine-tuned weights exist"
         }
       ],
+      "comparison_note": "Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. A camera-pose proxy forward-dynamics target export now passes the contract audit and schema-only packer smoke; true Cosmos3-Super fine-tuning is still not launched until the pipeline-loaded packer check and one-sample overfit exist."
     }
   ],
   "model_group_reading_notes": [
     "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
     "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
     "Cosmos3-Nano has a 128-episode future-window compatibility package.",
+    "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a camera-pose forward-dynamics contract audit; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist."
   ],
   "pending": [
     "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
+    "Promote Cosmos3 from Nano compatibility, Super base-weight evaluation, and the camera-pose forward-dynamics contract to true fine-tuning only after the pipeline-loaded packer check and one-sample overfit produce new weights."
   ]
 }

metrics/project_packet.json CHANGED Viewed

@@ -41,7 +41,7 @@
         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
-      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, and the Cosmos3-Super base-weight Reasoner evaluation are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

         "docs/data/scope_claims_audit.json",
         "docs/data/website_integrity.json"
       ],
+      "readout": "The project status table and roadmap give the compact current-state summary. Single-episode task engineering, metrics, visualizations, public website integrity, mirror parity, same-split 128-episode baselines, the final selected-episode Qwen3-Omni diagnostic result, the Cosmos3-Nano compatibility package, the Cosmos3-Super base-weight Reasoner evaluation, and the Cosmos3-Super camera-pose forward-dynamics contract audit plus schema-only packer smoke are implemented; stronger action/subtask and real Cosmos fine-tuned model quality remain follow-ups."
     },
     {
       "step": 2,

metrics/project_status.json CHANGED Viewed

@@ -119,7 +119,7 @@
         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
-      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package plus a verified Cosmos3-Super base-weight Reasoner evaluation; OpenVLA/openpi/GR00T are policy candidates after action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
@@ -244,6 +244,18 @@
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
@@ -276,11 +288,11 @@
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
-    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, and Cosmos3-Super Reasoner is verified as a base-weight evaluation; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
-    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, and policy models such as OpenVLA/openpi/GR00T wait for action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

         "FOUNDATION_MODEL_PLAN.md",
         "docs/data/foundation_model_plan.json"
       ],
+      "readout": "Qwen3-Omni remains the first trainable held-out LoRA baseline; Cosmos 3 is now represented by a verified Cosmos3-Nano future-window compatibility package, a verified Cosmos3-Super base-weight Reasoner evaluation, and a Cosmos3-Super camera-pose proxy forward-dynamics contract audit plus schema-only packer smoke. The current target supports vision-velocity training under action conditioning, not supervised action-token prediction; OpenVLA/openpi/GR00T are policy candidates after robot-compatible action targets are explicit."
     },
     {
       "area": "Omni model extension contract",
       ],
       "readout": "Cosmos3-Super Reasoner now has a public-safe verified 448-window held-out evaluation on the same structured JSON task as Qwen3. It uses staged nv-community/Cosmos3-Super base weights through an 8-GPU vLLM server, not fine-tuned weights: JSON validity 0.5112, action macro-F1 0.0008, transition accuracy 0.3683, contact accuracy 0.3214, and object micro-F1 0.1370."
     },
+    {
+      "area": "Cosmos3-Super action-target contract",
+      "status": "ready_for_forward_dynamics_trainer_implementation",
+      "evidence": [
+        "scripts/omni/export_cosmos3_camera_pose_targets.py",
+        "scripts/omni/pack_cosmos3_super_action_batch.py",
+        "results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/target_manifest.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json",
+        "results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json"
+      ],
+      "readout": "The selected 128-episode JSONL is augmented with 3,808/3,808 valid camera_pose proxy cosmos_action_target records from SLAM pose deltas. The schema-only packer smoke confirms the current forward_dynamics target should supervise noisy vision tokens under camera-pose conditioning; it does not supervise preds_action. Remaining work is a pipeline-loaded packer check, one-sample forward-dynamics overfit, and a separate policy/inverse target export before claiming action-token prediction."
+    },
     {
       "area": "Raw Xperience-10M redistribution",
       "status": "not_included",
     "Use docs/data/omni_model_comparison.json to compare both views: the single-episode/128-baseline/model-branch result layers and the model-family grouping for task heads, Qwen3-Omni LoRA, Cosmos3-Nano, and Cosmos3-Super.",
     "Use docs/data/omni_finetune_verified_result.json and the latest verified_public final Qwen package for current held-out results.",
     "The 128-episode aligned simple/NN baselines use metadata/text features from the derived Qwen JSONL export; they align the split and task ids but do not replace raw-modality baselines for trajectory, retrieval, reconstruction, or misalignment tasks.",
+    "The Cosmos3-Nano future-window branch is verified as a compatibility adapter result, Cosmos3-Super Reasoner is verified as a base-weight evaluation, and Cosmos3-Super camera-pose forward-dynamics targets now pass the contract audit plus a schema-only packer smoke; one-episode Cosmos fine-tuning and full Cosmos adapter/diffusion-weight fine-tuning remain pending, so no Cosmos weight repo should be published yet.",
     "The current reconstruction task reconstructs feature vectors, not pixel-depth, mesh, NeRF, or Gaussian reconstruction.",
     "Audio is one of the synchronized source modalities in the current task representation.",
     "The audio ablation report compares audio/no-audio variants across all 12 task contracts in results/audio_ablation/.",
+    "Foundation-model selection is explicit: Qwen3-Omni is the immediate trainable pilot, Cosmos 3 is the first world-model branch, Cosmos3-Super has a camera-pose proxy forward-dynamics contract ready for trainer implementation, and policy models such as OpenVLA/openpi/GR00T wait for robot-compatible action-target conversion.",
     "Future model branches should be added through the backbone registry and verified package contract, not as one-off result folders with incompatible metrics or publication rules.",
     "The Xperience Embodied Foundation Model is a future native-pretraining goal, not a completed model or current benchmark."
   ]

metrics/research_roadmap.json CHANGED Viewed

@@ -133,7 +133,7 @@
         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

         "docs/data/foundation_model_plan.json",
         "research_roadmap_interactive.json"
       ],
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets."
     },
     {
       "id": "robustness_run_64_128_episode",

metrics/research_roadmap_interactive.json CHANGED Viewed

@@ -2369,7 +2369,7 @@
       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
-      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch; VLA/policy models wait for explicit action targets.",
       "stage": "omni",
       "status": "next"
     },

       "entry_condition": "The selected episodes are prepared or a 3-8 episode dry run is available for preprocessing checks.",
       "id": "foundation_model_selection_matrix",
       "name": "Foundation-Model Selection Matrix",
+      "reader_takeaway": "Qwen3-Omni remains the first trainable held-out pilot; Cosmos 3 is the first world-model branch. Cosmos3-Super now has camera-pose proxy forward-dynamics targets ready for trainer implementation, while VLA/policy models wait for robot-compatible action targets.",
       "stage": "omni",
       "status": "next"
     },

metrics/website_integrity.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-07T15:47:32+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -75,7 +75,7 @@
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
-      "evidence_index": 90477
     },
     {
       "name": "project_status_links_json",
@@ -153,8 +153,8 @@
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
-      "protocol_index": 87160,
-      "evidence_index": 90477
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -292,7 +292,7 @@
     },
     {
       "path": "data/mirror_parity.json",
-      "bytes": 410374,
       "top_level_type": "dict"
     },
     {
@@ -302,12 +302,12 @@
     },
     {
       "path": "data/omni_finetune_verified_result.json",
-      "bytes": 3628,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
-      "bytes": 48296,
       "top_level_type": "dict"
     },
     {
@@ -322,12 +322,12 @@
     },
     {
       "path": "data/project_packet.json",
-      "bytes": 8005,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
-      "bytes": 16455,
       "top_level_type": "dict"
     },
     {
@@ -367,12 +367,12 @@
     },
     {
       "path": "data/research_roadmap.json",
-      "bytes": 10133,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
-      "bytes": 143560,
       "top_level_type": "dict"
     },
     {

 {
   "status": "pass",
+  "generated_at_utc": "2026-06-07T17:31:44+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
       "overview_index": 67412,
+      "evidence_index": 90659
     },
     {
       "name": "project_status_links_json",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
       "overview_index": 67412,
+      "protocol_index": 87218,
+      "evidence_index": 90659
     },
     {
       "name": "evaluation_protocol_links_json",
     },
     {
       "path": "data/mirror_parity.json",
+      "bytes": 345072,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/omni_finetune_verified_result.json",
+      "bytes": 3768,
       "top_level_type": "dict"
     },
     {
       "path": "data/omni_model_comparison.json",
+      "bytes": 51589,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/project_packet.json",
+      "bytes": 8098,
       "top_level_type": "dict"
     },
     {
       "path": "data/project_status.json",
+      "bytes": 18062,
       "top_level_type": "dict"
     },
     {
     },
     {
       "path": "data/research_roadmap.json",
+      "bytes": 10246,
       "top_level_type": "dict"
     },
     {
       "path": "data/research_roadmap_interactive.json",
+      "bytes": 143673,
       "top_level_type": "dict"
     },
     {

results/omni_finetune/OMNI_MODEL_COMPARISON.md CHANGED Viewed

@@ -1,6 +1,6 @@
 # Omni Model Comparison
-Generated: `2026-06-07T15:34:51+00:00`
 Compare only rows with the same scope and target. Single-episode raw-feature metrics, 128-episode metadata baselines, Qwen3 structured JSON metrics, and the two Cosmos3 targets answer different questions: Nano future-window retrieval versus Super structured JSON Reasoner evaluation.
@@ -16,7 +16,7 @@ Read the three rows this way:
 - Version 1 is the public-sample 12-task harness with minimal and neural heads.
 - Version 2 is the selected 128-episode same-split simple/NN baseline alignment.
-- Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation rather than a new fine-tuned weight release.
 ## Model-Family Grouped View
@@ -24,7 +24,7 @@ Read the three rows this way:
 - Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.
 - Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.
 - Cosmos3-Nano has a 128-episode future-window compatibility package.
-- Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a training-readiness probe; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist.
 ### Minimal and Neural Task Heads
@@ -64,7 +64,7 @@ The current 128-episode Cosmos result is a public-safe future-window compatibili
 ### Cosmos3-Super Reasoner
-Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. The readiness probe records why true Cosmos3-Super fine-tuning is not launched yet.
 - Weight repo policy: none for this run; staged base weights only, no new fine-tuned weights
@@ -72,6 +72,8 @@ Cosmos3-Super is now represented by a verified 448-window held-out Reasoner eval
 | --- | --- | --- | --- | --- | --- |
 | 1 episode | not_run | Cosmos3-Super One-Episode Fine-Tune |  |  |  |
 | readiness | blocked_until_trainer_implemented | Cosmos3-Super Training Readiness Probe | 3808 windows/samples | diffusers_runtime_supported=True, chat_sft_supported=False, weights_updated=False | `results/omni_finetune/xperience10m_cosmos3_super_training_readiness_20260607/training_readiness.json` |
 | 128 episode | verified current | Cosmos3-Super Reasoner | 119 episodes, 3808 windows/samples, 448 eval | json_validity_rate=0.5112, action_macro_f1=0.0008, transition_accuracy=0.3683, contact_accuracy=0.3214 | `results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/verified_result_summary.json` |
 ## 128-Episode Task Baselines
@@ -105,4 +107,4 @@ Cosmos3-Super is now represented by a verified 448-window held-out Reasoner eval
 ## Pending
 - Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.
-- Promote Cosmos3 from Nano compatibility and Super base-weight evaluation to true fine-tuning only after a dedicated Cosmos diffusion/action target packer and supervised loss produce new weights.

 # Omni Model Comparison
+Generated: `2026-06-07T17:29:16+00:00`
 Compare only rows with the same scope and target. Single-episode raw-feature metrics, 128-episode metadata baselines, Qwen3 structured JSON metrics, and the two Cosmos3 targets answer different questions: Nano future-window retrieval versus Super structured JSON Reasoner evaluation.
 - Version 1 is the public-sample 12-task harness with minimal and neural heads.
 - Version 2 is the selected 128-episode same-split simple/NN baseline alignment.
+- Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation; Cosmos3-Super now has a camera-pose forward-dynamics contract audit and schema-only packer smoke, but no new fine-tuned weight release.
 ## Model-Family Grouped View
 - Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.
 - Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.
 - Cosmos3-Nano has a 128-episode future-window compatibility package.
+- Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a camera-pose forward-dynamics contract audit; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist.
 ### Minimal and Neural Task Heads
 ### Cosmos3-Super Reasoner
+Cosmos3-Super is now represented by a verified 448-window held-out Reasoner evaluation on the same JSON task as Qwen3. It uses staged base weights through vLLM, so it is a model-branch diagnostic, not a weight release. A camera-pose proxy forward-dynamics target export now passes the contract audit and schema-only packer smoke; true Cosmos3-Super fine-tuning is still not launched until the pipeline-loaded packer check and one-sample overfit exist.
 - Weight repo policy: none for this run; staged base weights only, no new fine-tuned weights
 | --- | --- | --- | --- | --- | --- |
 | 1 episode | not_run | Cosmos3-Super One-Episode Fine-Tune |  |  |  |
 | readiness | blocked_until_trainer_implemented | Cosmos3-Super Training Readiness Probe | 3808 windows/samples | diffusers_runtime_supported=True, chat_sft_supported=False, weights_updated=False | `results/omni_finetune/xperience10m_cosmos3_super_training_readiness_20260607/training_readiness.json` |
+| action target contract | ready_for_forward_dynamics_trainer | Cosmos3-Super Camera-Pose Target Audit | 3808 windows/samples | domain_name=camera_pose, raw_action_dim=9, mode=forward_dynamics, valid_action_targets=3808, weights_updated=False | `results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_camera_pose_20260608/training_contract_audit.json` |
+| batch packer | pass | Cosmos3-Super Action Batch Packer Smoke | 1 windows/samples | mode=forward_dynamics, loss_surface=vision_velocity_conditioned_on_camera_pose, pipeline_loaded=False, weights_updated=False | `results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json` |
 | 128 episode | verified current | Cosmos3-Super Reasoner | 119 episodes, 3808 windows/samples, 448 eval | json_validity_rate=0.5112, action_macro_f1=0.0008, transition_accuracy=0.3683, contact_accuracy=0.3214 | `results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/verified_result_summary.json` |
 ## 128-Episode Task Baselines
 ## Pending
 - Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.
+- Promote Cosmos3 from Nano compatibility, Super base-weight evaluation, and the camera-pose forward-dynamics contract to true fine-tuning only after the pipeline-loaded packer check and one-sample overfit produce new weights.

results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/RUN_REPORT.md ADDED Viewed

	@@ -0,0 +1,19 @@

+# Cosmos3-Super Action Batch Packer
+- Run id: `xperience10m_cosmos3_super_action_packer_schema_smoke_20260608`
+- Row: `27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4:qa:0`
+- Mode: `forward_dynamics`
+- Domain: `camera_pose`
+- Raw action shape: `[8, 9]`
+- Pipeline loaded: `False`
+- Status: `pass`
+## Loss Surface
+- `vision_velocity_conditioned_on_camera_pose`
+- Cosmos3 forward_dynamics consumes raw_actions as conditioning and predicts noisy vision tokens. It does not supervise preds_action for this target mode.
+## Next Step
+- Implement the one-sample overfit with a vision velocity/rectified-flow loss under camera-pose action conditioning.
+- Add a separate policy or inverse-dynamics target export before claiming supervised action-token prediction.

results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/packer_summary.json ADDED Viewed

	@@ -0,0 +1,136 @@

+{
+  "run_id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608",
+  "run_kind": "cosmos3_super_action_batch_packer",
+  "started_at_unix": 1780852840.3893492,
+  "finished_at_unix": 1780852842.8621027,
+  "elapsed_seconds": 2.4727537631988525,
+  "dataset_jsonl": "/home/cy/Ropedia/ropedia-episode-task-suite/results/omni_finetune/xperience10m_cosmos3_camera_pose_targets_20260608/dataset_with_cosmos_actions.jsonl",
+  "backbone_config": "/home/cy/Ropedia/ropedia-episode-task-suite/configs/omni_backbones/cosmos3_super_reasoner.json",
+  "backbone": {
+    "id": "cosmos3_super_reasoner",
+    "display_name": "Cosmos3-Super Reasoner",
+    "status": "implemented",
+    "model_family": "Cosmos3 / physical-world foundation models",
+    "default_model_id": "nv-community/Cosmos3-Super",
+    "local_model_env": "COSMOS3_SUPER_MODEL_DIR",
+    "dataset_contract": "xperience10m_episode_json_qa_v1",
+    "training_objective": "zero_shot_structured_episode_understanding_json_qa_via_vllm_reasoner",
+    "split_policy": {
+      "unit": "episode",
+      "default_counts": {
+        "train": 96,
+        "val": 16,
+        "test": 16
+      },
+      "leakage_guard": "uses the same 96/16/16 selected episode split as the Qwen3-Omni LoRA branch; no Super weights are updated"
+    },
+    "modalities": {
+      "direct_inputs": [
+        "multi-camera rendered mosaic video",
+        "language prompt and label options"
+      ],
+      "conditioning_inputs": [
+        "prompt-side task schema and episode/window metadata"
+      ],
+      "targets": [
+        "structured action/subtask/contact/transition/object JSON"
+      ],
+      "excluded_inputs": [
+        "visualization.rrd",
+        "raw annotation HDF5",
+        "audio in the current vLLM Reasoner path"
+      ]
+    },
+    "entrypoints": {
+      "selection_manifest": "scripts/omni/build_selection_episode_manifest.py",
+      "export": "scripts/omni/parallel_export_qwen3_omni_action_dataset.py",
+      "neutral_index": "scripts/omni/export_model_neutral_window_index.py",
+      "action_target_export": "scripts/omni/export_cosmos3_camera_pose_targets.py",
+      "action_batch_packer": "scripts/omni/pack_cosmos3_super_action_batch.py",
+      "train": "",
+      "train_contract_audit": "scripts/omni/audit_cosmos3_super_training_contract.py",
+      "train_probe": "scripts/omni/probe_cosmos3_super_training_readiness.py",
+      "eval": "scripts/omni/eval_cosmos3_super_reasoner.py",
+      "launcher": "scripts/omni/run_cosmos3_super_reasoner_eval.sh",
+      "validate": "scripts/omni/validate_omni_finetune_run.py"
+    },
+    "primary_metrics": [
+      "json_validity_rate",
+      "action_macro_f1",
+      "subtask_accuracy",
+      "transition_accuracy",
+      "next_action_accuracy",
+      "contact_accuracy",
+      "object_micro_f1",
+      "held_out_episode_count"
+    ],
+    "artifact_contract": {
+      "checkpoint_gate": "base_weight_vllm_reasoner_setup_metadata",
+      "required_eval_files": [
+        "metrics.json",
+        "predictions.jsonl",
+        "predictions.csv",
+        "per_class_metrics.csv",
+        "confusion_matrix.csv",
+        "server_info.json",
+        "RUN_REPORT.md"
+      ],
+      "required_training_files": [
+        "training_metadata.json",
+        "progress.jsonl"
+      ],
+      "public_package_allowed": [
+        "metrics",
+        "predictions",
+        "confusion matrices",
+        "run reports",
+        "server/model setup metadata",
+        "episode and dataset manifests",
+        "validation summaries"
+      ],
+      "public_package_forbidden": [
+        "raw MP4",
+        "annotation HDF5",
+        "Rerun RRD",
+        "base-model weights",
+        "fine-tuned weights",
+        "checkpoints",
+        "large archives"
+      ]
+    },
+    "extension_requirements": [
+      "This branch evaluates staged Cosmos3-Super Reasoner base weights through vLLM on the 128-episode held-out JSON task; it does not fine-tune or release new Cosmos weights.",
+      "Run scripts/omni/probe_cosmos3_super_training_readiness.py before any Cosmos3-Super adapter launch; the probe must have no blockers before train can be filled.",
+      "Create a separate Cosmos3-Super adapter/model repository only after a real fine-tuning run produces new adapter or checkpoint weights.",
+      "Keep it separate from the Cosmos3-Nano future-window compatibility branch, which answers a different world-model retrieval target."
+    ]
+  },
+  "status": "pass",
+  "row_contract": {
+    "row_id": "27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4:qa:0",
+    "episode_id": "27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4",
+    "split": "train",
+    "target_key": "cosmos_action_target",
+    "mode": "forward_dynamics",
+    "domain_name": "camera_pose",
+    "chunk_size": 8,
+    "raw_action_dim": 9,
+    "raw_actions_shape": [
+      8,
+      9
+    ],
+    "video_path": "/home/cy/Ropedia/ropedia-episode-task-suite/results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/shards/shard_00/media/27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4/27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4_w00000_ctx0_119_mosaic.mp4",
+    "video_path_exists": true,
+    "loss_surface": "vision_velocity_conditioned_on_camera_pose",
+    "action_loss_expected": false,
+    "interpretation": "Cosmos3 forward_dynamics consumes raw_actions as conditioning and predicts noisy vision tokens. It does not supervise preds_action for this target mode.",
+    "issues": []
+  },
+  "pack_result": {
+    "status": "schema_ready_pipeline_not_loaded",
+    "pipeline_loaded": false,
+    "loss_surface": "vision_velocity_conditioned_on_camera_pose",
+    "action_loss_expected": false
+  },
+  "weights_updated": false
+}

results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/progress.jsonl ADDED Viewed

	@@ -0,0 +1,3 @@

+{"event": "start", "run_id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608", "time": 1780852840.3893492}
+{"event": "row_selected", "row_id": "27c9fc42-2bb4-4737-b09c-08d2dd88aed4__ep4:qa:0", "time": 1780852842.8619707}
+{"event": "complete", "status": "pass", "time": 1780852842.8629975}

results/omni_finetune/xperience10m_cosmos3_super_action_packer_schema_smoke_20260608/training_metadata.json ADDED Viewed

	@@ -0,0 +1,8 @@

+{
+  "run_id": "xperience10m_cosmos3_super_action_packer_schema_smoke_20260608",
+  "run_kind": "cosmos3_super_action_batch_packer",
+  "weights_updated": false,
+  "checkpoint_dir": null,
+  "status": "pass",
+  "loss_surface": "vision_velocity_conditioned_on_camera_pose"
+}

results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/RUN_REPORT.md ADDED Viewed

	@@ -0,0 +1,35 @@

+# Cosmos3-Super Training Contract Audit
+- Run id: `xperience10m_cosmos3_super_training_contract_audit_local`
+- Dataset: `/Users/chaoyue/Documents/Codex/2026-05-29/i-am-learning-this-dataset-https/working_repo_copy/results/omni_finetune/dataset.jsonl`
+- Rows: `128`
+- Rows with Cosmos action targets: `0`
+- Valid Cosmos action targets: `0`
+- Status: `blocked_missing_cosmos_action_targets`
+- Weights updated: `False`
+## Blockers
+- dataset has no cosmos_action_target/cosmos3_action_target/action_target records; semantic JSON labels cannot be used as Cosmos continuous action latents
+## Required Target Schema
+```json
+{
+  "cosmos_action_target": {
+    "mode": "policy|forward_dynamics|inverse_dynamics",
+    "domain_name": "one Cosmos3 embodiment domain supported by CosmosActionCondition",
+    "chunk_size": "positive integer action transition count",
+    "raw_actions": "required for forward_dynamics; list[list[float]] with shape [T, raw_action_dim]",
+    "video": "required for inverse_dynamics, or image/video conditioning for policy and forward_dynamics",
+    "resolution_tier": "optional; one of 256, 480, 704, 720",
+    "view_point": "optional; ego_view|third_person_view|wrist_view|concat_view"
+  }
+}
+```
+## Next Steps
+- Export Cosmos-native action targets from Xperience annotations or mocap/pose/contact signals into the required cosmos_action_target schema.
+- Implement a one-sample batch packer that calls Cosmos3OmniPipeline.prepare_latents and the static segment helpers, then computes MSE/rectified-flow loss over preds_action for noisy action tokens.
+- Run a one-episode overfit before scheduling a 96/16/16 Super LoRA run; only publish a Cosmos model repo after new adapter/checkpoint weights exist.

results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/progress.jsonl ADDED Viewed

	@@ -0,0 +1,3 @@

+{"event": "start", "run_id": "xperience10m_cosmos3_super_training_contract_audit_local", "time": 1780849944.267908}
+{"event": "dataset_loaded", "rows": 128, "time": 1780849944.278147}
+{"event": "complete", "status": "blocked_missing_cosmos_action_targets", "time": 1780849944.2802079}

results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/training_contract_audit.json ADDED Viewed

	@@ -0,0 +1,78 @@

+{
+  "run_id": "xperience10m_cosmos3_super_training_contract_audit_local",
+  "run_kind": "cosmos3_super_training_contract_audit",
+  "started_at_unix": 1780849944.267908,
+  "finished_at_unix": 1780849944.279339,
+  "elapsed_seconds": 0.011430978775024414,
+  "workspace": "/Users/chaoyue/Documents/Codex/2026-05-29/i-am-learning-this-dataset-https/working_repo_copy",
+  "dataset_jsonl": "/Users/chaoyue/Documents/Codex/2026-05-29/i-am-learning-this-dataset-https/working_repo_copy/results/omni_finetune/dataset.jsonl",
+  "sample_limit": 0,
+  "backbone_config": "/Users/chaoyue/Documents/Codex/2026-05-29/i-am-learning-this-dataset-https/working_repo_copy/configs/omni_backbones/cosmos3_super_reasoner.json",
+  "backbone": {
+    "id": "cosmos3_super_reasoner",
+    "display_name": "Cosmos3-Super Reasoner",
+    "training_objective": "zero_shot_structured_episode_understanding_json_qa_via_vllm_reasoner"
+  },
+  "model": {
+    "provided": false
+  },
+  "dataset": {
+    "num_rows": 128,
+    "split_counts": {
+      "train": 128
+    },
+    "episode_split_counts": {
+      "train": 1
+    },
+    "rows_with_video": 128,
+    "missing_json_answer": 0,
+    "missing_json_fields": {},
+    "rows_with_action_target": 0,
+    "valid_action_targets": 0,
+    "target_key_counts": {},
+    "target_mode_counts": {},
+    "target_issue_counts": {},
+    "target_issue_examples": []
+  },
+  "decision": {
+    "status": "blocked_missing_cosmos_action_targets",
+    "weights_updated": false,
+    "blockers": [
+      "dataset has no cosmos_action_target/cosmos3_action_target/action_target records; semantic JSON labels cannot be used as Cosmos continuous action latents"
+    ],
+    "warnings": [
+      "model_dir not provided; model action_gen/action_dim could not be verified"
+    ],
+    "required_target_schema": {
+      "cosmos_action_target": {
+        "mode": "policy|forward_dynamics|inverse_dynamics",
+        "domain_name": "one Cosmos3 embodiment domain supported by CosmosActionCondition",
+        "chunk_size": "positive integer action transition count",
+        "raw_actions": "required for forward_dynamics; list[list[float]] with shape [T, raw_action_dim]",
+        "video": "required for inverse_dynamics, or image/video conditioning for policy and forward_dynamics",
+        "resolution_tier": "optional; one of 256, 480, 704, 720",
+        "view_point": "optional; ego_view|third_person_view|wrist_view|concat_view"
+      }
+    },
+    "trainer_contract": {
+      "diffusers_classes": [
+        "Cosmos3OmniPipeline",
+        "Cosmos3OmniTransformer",
+        "CosmosActionCondition"
+      ],
+      "packing_helpers": [
+        "Cosmos3OmniPipeline.prepare_latents",
+        "Cosmos3OmniPipeline._prepare_text_segment",
+        "Cosmos3OmniPipeline._prepare_vision_segment",
+        "Cosmos3OmniPipeline._prepare_action_segment"
+      ],
+      "forward_outputs": "Cosmos3OmniTransformer.forward returns (preds_vision, preds_sound, preds_action); action LoRA needs supervised loss against raw continuous action tokens, not JSON strings.",
+      "lora_targets": "use checkpoint-declared q_proj_moe_gen,k_proj_moe_gen,v_proj_moe_gen,o_proj_moe_gen unless a new audited config overrides them"
+    },
+    "next_steps": [
+      "Export Cosmos-native action targets from Xperience annotations or mocap/pose/contact signals into the required cosmos_action_target schema.",
+      "Implement a one-sample batch packer that calls Cosmos3OmniPipeline.prepare_latents and the static segment helpers, then computes MSE/rectified-flow loss over preds_action for noisy action tokens.",
+      "Run a one-episode overfit before scheduling a 96/16/16 Super LoRA run; only publish a Cosmos model repo after new adapter/checkpoint weights exist."
+    ]
+  }
+}

results/omni_finetune/xperience10m_cosmos3_super_training_contract_audit_local/training_metadata.json ADDED Viewed

	@@ -0,0 +1,47 @@

+{
+  "run_id": "xperience10m_cosmos3_super_training_contract_audit_local",
+  "run_kind": "cosmos3_super_training_contract_audit",
+  "weights_updated": false,
+  "checkpoint_dir": null,
+  "decision": {
+    "status": "blocked_missing_cosmos_action_targets",
+    "weights_updated": false,
+    "blockers": [
+      "dataset has no cosmos_action_target/cosmos3_action_target/action_target records; semantic JSON labels cannot be used as Cosmos continuous action latents"
+    ],
+    "warnings": [
+      "model_dir not provided; model action_gen/action_dim could not be verified"
+    ],
+    "required_target_schema": {
+      "cosmos_action_target": {
+        "mode": "policy|forward_dynamics|inverse_dynamics",
+        "domain_name": "one Cosmos3 embodiment domain supported by CosmosActionCondition",
+        "chunk_size": "positive integer action transition count",
+        "raw_actions": "required for forward_dynamics; list[list[float]] with shape [T, raw_action_dim]",
+        "video": "required for inverse_dynamics, or image/video conditioning for policy and forward_dynamics",
+        "resolution_tier": "optional; one of 256, 480, 704, 720",
+        "view_point": "optional; ego_view|third_person_view|wrist_view|concat_view"
+      }
+    },
+    "trainer_contract": {
+      "diffusers_classes": [
+        "Cosmos3OmniPipeline",
+        "Cosmos3OmniTransformer",
+        "CosmosActionCondition"
+      ],
+      "packing_helpers": [
+        "Cosmos3OmniPipeline.prepare_latents",
+        "Cosmos3OmniPipeline._prepare_text_segment",
+        "Cosmos3OmniPipeline._prepare_vision_segment",
+        "Cosmos3OmniPipeline._prepare_action_segment"
+      ],
+      "forward_outputs": "Cosmos3OmniTransformer.forward returns (preds_vision, preds_sound, preds_action); action LoRA needs supervised loss against raw continuous action tokens, not JSON strings.",
+      "lora_targets": "use checkpoint-declared q_proj_moe_gen,k_proj_moe_gen,v_proj_moe_gen,o_proj_moe_gen unless a new audited config overrides them"
+    },
+    "next_steps": [
+      "Export Cosmos-native action targets from Xperience annotations or mocap/pose/contact signals into the required cosmos_action_target schema.",
+      "Implement a one-sample batch packer that calls Cosmos3OmniPipeline.prepare_latents and the static segment helpers, then computes MSE/rectified-flow loss over preds_action for noisy action tokens.",
+      "Run a one-episode overfit before scheduling a 96/16/16 Super LoRA run; only publish a Cosmos model repo after new adapter/checkpoint weights exist."
+    ]
+  }
+}

scripts/omni/audit_cosmos3_super_training_contract.py ADDED Viewed

	@@ -0,0 +1,406 @@

+#!/usr/bin/env python3
+"""Audit whether a dataset can drive real Cosmos3-Super action fine-tuning.
+The existing Cosmos3-Super Reasoner run evaluates base weights on structured
+JSON QA. A true Cosmos3 Diffusers fine-tune is a different contract: the
+transformer action path predicts continuous embodiment-domain action vectors,
+not semantic JSON labels. This guard makes that distinction explicit and fails
+closed until the exported Xperience-10M windows contain Cosmos-native action
+targets.
+"""
+from __future__ import annotations
+import argparse
+import json
+import math
+import time
+from collections import Counter
+from pathlib import Path
+from typing import Any
+from qwen3_omni_dataset_utils import load_jsonl
+REQUIRED_JSON_QA_FIELDS = {
+    "action",
+    "subtask",
+    "objects",
+    "contact",
+    "transition",
+    "next_action",
+    "evidence_window",
+}
+ACTION_TARGET_KEYS = (
+    "cosmos_action_target",
+    "cosmos3_action_target",
+    "cosmos_action_condition",
+    "action_target",
+)
+REQUIRED_ACTION_TARGET_FIELDS = {
+    "mode",
+    "domain_name",
+    "chunk_size",
+}
+ACTION_MODES = {"policy", "forward_dynamics", "inverse_dynamics"}
+REQUIRED_SCHEMA = {
+    "cosmos_action_target": {
+        "mode": "policy|forward_dynamics|inverse_dynamics",
+        "domain_name": "one Cosmos3 embodiment domain supported by CosmosActionCondition",
+        "chunk_size": "positive integer action transition count",
+        "raw_actions": "required for forward_dynamics; list[list[float]] with shape [T, raw_action_dim]",
+        "video": "required for inverse_dynamics, or image/video conditioning for policy and forward_dynamics",
+        "resolution_tier": "optional; one of 256, 480, 704, 720",
+        "view_point": "optional; ego_view|third_person_view|wrist_view|concat_view",
+    }
+}
+def parse_args() -> argparse.Namespace:
+    workspace_default = Path(__file__).resolve().parents[2]
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--workspace", type=Path, default=workspace_default)
+    parser.add_argument("--dataset-jsonl", type=Path, required=True)
+    parser.add_argument("--model-dir", type=Path)
+    parser.add_argument(
+        "--backbone-config",
+        type=Path,
+        default=workspace_default / "configs" / "omni_backbones" / "cosmos3_super_reasoner.json",
+    )
+    parser.add_argument("--run-id", default="xperience10m_cosmos3_super_training_contract_audit")
+    parser.add_argument("--output-dir", type=Path)
+    parser.add_argument("--sample-limit", type=int, default=0)
+    parser.add_argument(
+        "--require-trainable",
+        action="store_true",
+        help="Exit non-zero unless the dataset/model contract is ready for a real trainer launch.",
+    )
+    return parser.parse_args()
+def read_json(path: Path | None) -> dict[str, Any]:
+    if path is None or not path.exists():
+        return {}
+    return json.loads(path.read_text(encoding="utf-8"))
+def write_json(path: Path, payload: dict[str, Any]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(json.dumps(payload, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+def append_jsonl(path: Path, payload: dict[str, Any]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("a", encoding="utf-8") as handle:
+        handle.write(json.dumps(payload, sort_keys=True, ensure_ascii=False) + "\n")
+def numeric_matrix(value: Any) -> tuple[bool, tuple[int, int] | None]:
+    if not isinstance(value, list) or not value:
+        return False, None
+    width: int | None = None
+    for row in value:
+        if not isinstance(row, list) or not row:
+            return False, None
+        if width is None:
+            width = len(row)
+        elif len(row) != width:
+            return False, None
+        for item in row:
+            if not isinstance(item, (int, float)) or not math.isfinite(float(item)):
+                return False, None
+    return True, (len(value), int(width or 0))
+def find_action_target(row: dict[str, Any]) -> tuple[str | None, dict[str, Any] | None]:
+    for key in ACTION_TARGET_KEYS:
+        value = row.get(key)
+        if isinstance(value, dict):
+            return key, value
+    return None, None
+def media_has_video(row: dict[str, Any]) -> bool:
+    media = row.get("media") if isinstance(row.get("media"), dict) else {}
+    if media.get("mosaic_video_path") or row.get("primary_video_path"):
+        return True
+    video_paths = media.get("video_paths")
+    return isinstance(video_paths, list) and any(isinstance(item, dict) and item.get("path") for item in video_paths)
+def validate_action_target(target: dict[str, Any]) -> list[str]:
+    issues: list[str] = []
+    missing = sorted(field for field in REQUIRED_ACTION_TARGET_FIELDS if field not in target)
+    if missing:
+        issues.append(f"missing fields: {missing}")
+        return issues
+    mode = str(target.get("mode"))
+    if mode not in ACTION_MODES:
+        issues.append(f"unsupported mode: {mode!r}")
+    try:
+        chunk_size = int(target.get("chunk_size"))
+        if chunk_size < 1:
+            issues.append("chunk_size must be >= 1")
+    except Exception:
+        issues.append("chunk_size must be an integer")
+        chunk_size = 0
+    if not str(target.get("domain_name") or "").strip():
+        issues.append("domain_name is empty")
+    raw_actions = target.get("raw_actions")
+    if mode == "forward_dynamics":
+        ok, shape = numeric_matrix(raw_actions)
+        if not ok:
+            issues.append("forward_dynamics requires numeric raw_actions shaped [T, raw_action_dim]")
+        elif shape and shape[0] < 1:
+            issues.append("raw_actions must include at least one action row")
+    elif raw_actions is not None:
+        ok, _ = numeric_matrix(raw_actions)
+        if not ok:
+            issues.append("raw_actions is present but is not a numeric matrix")
+    return issues
+def model_summary(model_dir: Path | None) -> dict[str, Any]:
+    if model_dir is None:
+        return {"provided": False}
+    model_dir = model_dir.expanduser().resolve()
+    config = read_json(model_dir / "config.json")
+    transformer_config = read_json(model_dir / "transformer" / "config.json")
+    inner = ((config.get("model") or {}).get("config") or {})
+    return {
+        "provided": True,
+        "path": str(model_dir),
+        "exists": model_dir.exists(),
+        "model_type": config.get("model_type"),
+        "architectures": config.get("architectures"),
+        "pipeline_class": read_json(model_dir / "model_index.json").get("_class_name"),
+        "transformer_class": transformer_config.get("_class_name"),
+        "action_gen": transformer_config.get("action_gen", inner.get("action_gen")),
+        "action_dim": transformer_config.get("action_dim", inner.get("action_dim")),
+        "lora_enabled_default": inner.get("lora_enabled"),
+        "lora_rank_default": inner.get("lora_rank"),
+        "lora_alpha_default": inner.get("lora_alpha"),
+        "lora_target_modules_default": inner.get("lora_target_modules"),
+        "rectified_flow_training_config_keys": sorted(
+            ((inner.get("rectified_flow_training_config") or {}).keys())
+        ),
+    }
+def dataset_summary(rows: list[dict[str, Any]]) -> dict[str, Any]:
+    split_counts = Counter(str(row.get("split", "unspecified")) for row in rows)
+    episodes_by_split: dict[str, set[str]] = {}
+    missing_json_answer = 0
+    missing_json_fields = Counter()
+    rows_with_video = 0
+    rows_with_action_target = 0
+    valid_action_targets = 0
+    target_key_counts = Counter()
+    target_mode_counts = Counter()
+    target_issue_counts = Counter()
+    examples: list[dict[str, Any]] = []
+    for row in rows:
+        split = str(row.get("split", "unspecified"))
+        episodes_by_split.setdefault(split, set()).add(str(row.get("episode_id", "")))
+        answer = row.get("answer_json") if isinstance(row.get("answer_json"), dict) else {}
+        if not answer:
+            missing_json_answer += 1
+        for field in REQUIRED_JSON_QA_FIELDS:
+            if field not in answer:
+                missing_json_fields[field] += 1
+        if media_has_video(row):
+            rows_with_video += 1
+        key, target = find_action_target(row)
+        if target is None:
+            continue
+        rows_with_action_target += 1
+        target_key_counts[str(key)] += 1
+        target_mode_counts[str(target.get("mode", "missing"))] += 1
+        issues = validate_action_target(target)
+        if issues:
+            for issue in issues:
+                target_issue_counts[issue] += 1
+            if len(examples) < 5:
+                examples.append({"id": row.get("id"), "target_key": key, "issues": issues})
+        else:
+            valid_action_targets += 1
+    return {
+        "num_rows": len(rows),
+        "split_counts": dict(split_counts),
+        "episode_split_counts": {split: len(episodes) for split, episodes in sorted(episodes_by_split.items())},
+        "rows_with_video": rows_with_video,
+        "missing_json_answer": missing_json_answer,
+        "missing_json_fields": dict(missing_json_fields),
+        "rows_with_action_target": rows_with_action_target,
+        "valid_action_targets": valid_action_targets,
+        "target_key_counts": dict(target_key_counts),
+        "target_mode_counts": dict(target_mode_counts),
+        "target_issue_counts": dict(target_issue_counts),
+        "target_issue_examples": examples,
+    }
+def decide(dataset: dict[str, Any], model: dict[str, Any]) -> dict[str, Any]:
+    blockers: list[str] = []
+    warnings: list[str] = []
+    if dataset["num_rows"] <= 0:
+        blockers.append("dataset has zero rows")
+    if dataset["rows_with_video"] <= 0:
+        blockers.append("dataset has no video conditioning paths")
+    if dataset["missing_json_answer"] or dataset["missing_json_fields"]:
+        warnings.append("dataset is not a complete JSON QA export")
+    if model.get("provided"):
+        if not model.get("exists"):
+            blockers.append(f"model_dir does not exist: {model.get('path')}")
+        if model.get("model_type") != "cosmos3_omni":
+            warnings.append(f"model_type is not cosmos3_omni: {model.get('model_type')}")
+        if model.get("action_gen") is not True:
+            blockers.append("Cosmos3 transformer config does not advertise action_gen=True")
+        if not model.get("action_dim"):
+            blockers.append("Cosmos3 transformer config does not expose action_dim")
+    else:
+        warnings.append("model_dir not provided; model action_gen/action_dim could not be verified")
+    if dataset["rows_with_action_target"] <= 0:
+        blockers.append(
+            "dataset has no cosmos_action_target/cosmos3_action_target/action_target records; "
+            "semantic JSON labels cannot be used as Cosmos continuous action latents"
+        )
+    elif dataset["valid_action_targets"] != dataset["rows_with_action_target"]:
+        blockers.append(
+            "one or more action target records do not satisfy the CosmosActionCondition schema"
+        )
+    status = "ready_for_cosmos3_super_action_lora" if not blockers else "blocked_missing_cosmos_action_targets"
+    if not blockers and dataset.get("target_mode_counts") == {"forward_dynamics": dataset["rows_with_action_target"]}:
+        status = "ready_for_cosmos3_super_forward_dynamics_lora"
+    return {
+        "status": status,
+        "weights_updated": False,
+        "blockers": blockers,
+        "warnings": warnings,
+        "required_target_schema": REQUIRED_SCHEMA,
+        "trainer_contract": {
+            "diffusers_classes": [
+                "Cosmos3OmniPipeline",
+                "Cosmos3OmniTransformer",
+                "CosmosActionCondition",
+            ],
+            "packing_helpers": [
+                "Cosmos3OmniPipeline.prepare_latents",
+                "Cosmos3OmniPipeline._prepare_text_segment",
+                "Cosmos3OmniPipeline._prepare_vision_segment",
+                "Cosmos3OmniPipeline._prepare_action_segment",
+            ],
+            "forward_outputs": "Cosmos3OmniTransformer.forward returns (preds_vision, preds_sound, preds_action). The current camera_pose forward_dynamics target uses raw actions as conditioning and should supervise preds_vision; supervised preds_action needs policy or inverse_dynamics targets.",
+            "lora_targets": "use checkpoint-declared q_proj_moe_gen,k_proj_moe_gen,v_proj_moe_gen,o_proj_moe_gen unless a new audited config overrides them",
+        },
+        "next_steps": [
+            "Run the one-sample action batch packer that calls Cosmos3OmniPipeline.prepare_latents and the static segment helpers, then records whether the current target supervises vision or action tokens.",
+            "For the current camera_pose forward_dynamics target, implement a one-sample overfit with vision velocity/rectified-flow loss under action conditioning; add a policy/inverse target export before claiming supervised action-token prediction.",
+            "Run a one-episode overfit before scheduling a 96/16/16 Super LoRA run; only publish a Cosmos model repo after new adapter/checkpoint weights exist.",
+        ],
+    }
+def write_report(path: Path, payload: dict[str, Any]) -> None:
+    decision = payload["decision"]
+    lines = [
+        "# Cosmos3-Super Training Contract Audit",
+        "",
+        f"- Run id: `{payload['run_id']}`",
+        f"- Dataset: `{payload['dataset_jsonl']}`",
+        f"- Rows: `{payload['dataset']['num_rows']}`",
+        f"- Rows with Cosmos action targets: `{payload['dataset']['rows_with_action_target']}`",
+        f"- Valid Cosmos action targets: `{payload['dataset']['valid_action_targets']}`",
+        f"- Status: `{decision['status']}`",
+        f"- Weights updated: `{decision['weights_updated']}`",
+        "",
+        "## Blockers",
+        "",
+    ]
+    if decision["blockers"]:
+        lines.extend(f"- {item}" for item in decision["blockers"])
+    else:
+        lines.append("- None")
+    lines.extend(["", "## Required Target Schema", "", "```json", json.dumps(REQUIRED_SCHEMA, indent=2), "```", ""])
+    lines.extend(["## Next Steps", ""])
+    lines.extend(f"- {item}" for item in decision["next_steps"])
+    path.write_text("\n".join(lines) + "\n", encoding="utf-8")
+def main() -> int:
+    args = parse_args()
+    args.workspace = args.workspace.expanduser().resolve()
+    args.dataset_jsonl = args.dataset_jsonl.expanduser().resolve()
+    if args.model_dir is not None:
+        args.model_dir = args.model_dir.expanduser().resolve()
+    output_dir = args.output_dir or args.workspace / "results" / "omni_finetune" / args.run_id
+    output_dir = output_dir.expanduser().resolve()
+    output_dir.mkdir(parents=True, exist_ok=True)
+    progress_path = output_dir / "progress.jsonl"
+    started = time.time()
+    append_jsonl(progress_path, {"event": "start", "time": started, "run_id": args.run_id})
+    rows = load_jsonl(args.dataset_jsonl)
+    if args.sample_limit > 0:
+        rows = rows[: args.sample_limit]
+    append_jsonl(progress_path, {"event": "dataset_loaded", "time": time.time(), "rows": len(rows)})
+    dataset = dataset_summary(rows)
+    model = model_summary(args.model_dir)
+    backbone = read_json(args.backbone_config)
+    decision = decide(dataset, model)
+    payload = {
+        "run_id": args.run_id,
+        "run_kind": "cosmos3_super_training_contract_audit",
+        "started_at_unix": started,
+        "finished_at_unix": time.time(),
+        "elapsed_seconds": time.time() - started,
+        "workspace": str(args.workspace),
+        "dataset_jsonl": str(args.dataset_jsonl),
+        "sample_limit": args.sample_limit,
+        "backbone_config": str(args.backbone_config),
+        "backbone": {
+            "id": backbone.get("id"),
+            "display_name": backbone.get("display_name"),
+            "training_objective": backbone.get("training_objective"),
+        },
+        "model": model,
+        "dataset": dataset,
+        "decision": decision,
+    }
+    write_json(output_dir / "training_contract_audit.json", payload)
+    write_json(output_dir / "training_metadata.json", {
+        "run_id": args.run_id,
+        "run_kind": payload["run_kind"],
+        "weights_updated": False,
+        "checkpoint_dir": None,
+        "decision": decision,
+    })
+    write_report(output_dir / "RUN_REPORT.md", payload)
+    append_jsonl(progress_path, {"event": "complete", "time": time.time(), "status": decision["status"]})
+    print(json.dumps({"status": decision["status"], "output_dir": str(output_dir)}, indent=2))
+    ready_statuses = {
+        "ready_for_cosmos3_super_action_lora",
+        "ready_for_cosmos3_super_forward_dynamics_lora",
+    }
+    return 1 if args.require_trainable and decision["status"] not in ready_statuses else 0
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/build_omni_model_comparison.py CHANGED Viewed

@@ -315,8 +315,93 @@ def cosmos3_super_readiness_entry() -> dict[str, Any] | None:
         "weights": "none; readiness audit only, no adapter checkpoint",
         "interpretation": (
             "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and "
-            "the same JSON QA dataset are visible, but blocks true fine-tuning until "
-            "a Cosmos-specific diffusion/action target packer and supervised loss are implemented."
         ),
     }
@@ -344,6 +429,8 @@ def model_grouped_view(versions: list[dict[str, Any]]) -> list[dict[str, Any]]:
     cosmos_nano_branches = [branch for branch in branches if branch.get("backbone") == "cosmos_world_model"]
     cosmos_super_branches = [branch for branch in branches if branch.get("backbone") == "cosmos3_super_reasoner"]
     cosmos_super_readiness = cosmos3_super_readiness_entry()
     if qwen_branches:
         current_qwen = max(qwen_branches, key=lambda item: item.get("primary_metrics", {}).get("json_validity_rate") or -1)
         for branch in qwen_branches:
@@ -451,13 +538,17 @@ def model_grouped_view(versions: list[dict[str, Any]]) -> list[dict[str, Any]]:
                     ),
                 }
             ],
-            "readiness_runs": [cosmos_super_readiness] if cosmos_super_readiness else [],
             "multi_episode_128_runs": cosmos_super_branches,
             "comparison_note": (
                 "Cosmos3-Super is now represented by a verified 448-window held-out "
                 "Reasoner evaluation on the same JSON task as Qwen3. It uses staged base "
                 "weights through vLLM, so it is a model-branch diagnostic, not a weight release. "
-                "The readiness probe records why true Cosmos3-Super fine-tuning is not launched yet."
             ),
         },
     ]
@@ -481,7 +572,7 @@ def build_report() -> dict[str, Any]:
         "version_reading_notes": [
             "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
             "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
-            "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation rather than a new fine-tuned weight release.",
         ],
         "versions": versions,
         "model_groups": model_groups,
@@ -490,11 +581,11 @@ def build_report() -> dict[str, Any]:
             "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
             "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
             "Cosmos3-Nano has a 128-episode future-window compatibility package.",
-            "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a training-readiness probe; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist.",
         ],
         "pending": [
             "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
-            "Promote Cosmos3 from Nano compatibility and Super base-weight evaluation to true fine-tuning only after a dedicated Cosmos diffusion/action target packer and supervised loss produce new weights.",
         ],
     }
@@ -512,7 +603,7 @@ def entry_count_text(entry: dict[str, Any]) -> str:
     pieces = []
     for label, keys in (
         ("episodes", ("episodes", "dataset_episodes", "held_out_episode_count")),
-        ("windows/samples", ("windows", "rows", "dataset_samples", "eval_samples")),
         ("eval", ("eval_samples",)),
     ):
         value = next((counts.get(key) for key in keys if counts.get(key) is not None), None)
@@ -534,6 +625,12 @@ def entry_metric_text(entry: dict[str, Any]) -> str:
         "contact_accuracy",
         "accuracy",
         "macro_f1",
         "diffusers_runtime_supported",
         "chat_sft_supported",
         "weights_updated",
@@ -559,7 +656,7 @@ def append_model_group(lines: list[str], group: dict[str, Any]) -> None:
     for entry in group.get("one_episode_runs", []):
         rows.append(("1 episode", entry))
     for entry in group.get("readiness_runs", []):
-        rows.append(("readiness", entry))
     for entry in group.get("multi_episode_128_runs", []):
         rows.append(("128 episode", entry))
     for scope, entry in rows:

         "weights": "none; readiness audit only, no adapter checkpoint",
         "interpretation": (
             "This probe confirms the staged Cosmos3-Super Diffusers/GPU runtime and "
+            "the same JSON QA dataset are visible. It predates the camera-pose action-target "
+            "export, so use the 20260608 contract audit for the current trainer-readiness status."
+        ),
+    }
+def cosmos3_super_action_contract_entry() -> dict[str, Any] | None:
+    paths = sorted(
+        (ROOT / "results/omni_finetune").glob(
+            "xperience10m_cosmos3_super_training_contract_audit_*/training_contract_audit.json"
+        )
+    )
+    if not paths:
+        return None
+    payloads = [(path, load_json(path)) for path in paths]
+    path, payload = max(payloads, key=lambda item: item[1].get("finished_at_unix") or 0)
+    decision = payload.get("decision", {}) if isinstance(payload.get("decision"), dict) else {}
+    dataset = payload.get("dataset", {}) if isinstance(payload.get("dataset"), dict) else {}
+    target_modes = dataset.get("target_mode_counts", {}) if isinstance(dataset.get("target_mode_counts"), dict) else {}
+    only_forward_dynamics = set(target_modes) == {"forward_dynamics"}
+    return {
+        "id": payload.get("run_id", path.parent.name),
+        "title": "Cosmos3-Super Camera-Pose Target Audit",
+        "scope_label": "action target contract",
+        "scope": "selected 128-episode 96/16/16 dataset augmented with camera_pose proxy cosmos_action_target records",
+        "status": "ready_for_forward_dynamics_trainer" if only_forward_dynamics else "ready_for_action_lora_trainer" if decision.get("status") == "ready_for_cosmos3_super_action_lora" else decision.get("status", "unknown"),
+        "source": rel(path),
+        "split": "train/val/test by selected episode/session",
+        "counts": {
+            "dataset_samples": dataset.get("num_rows"),
+            "rows_with_action_target": dataset.get("rows_with_action_target"),
+            "valid_action_targets": dataset.get("valid_action_targets"),
+            "split_counts": dataset.get("split_counts"),
+            "episode_split_counts": dataset.get("episode_split_counts"),
+        },
+        "primary_metrics": {
+            "domain_name": "camera_pose",
+            "raw_action_dim": 9,
+            "mode": next(iter(target_modes), "forward_dynamics"),
+            "valid_action_targets": dataset.get("valid_action_targets"),
+            "weights_updated": decision.get("weights_updated"),
+        },
+        "weights": "none; action-target contract audit only, no adapter checkpoint",
+        "interpretation": (
+            "The selected dataset now has valid Cosmos3 camera_pose forward_dynamics targets "
+            "for an egocentric camera-motion proxy. These remove the target-schema blocker "
+            "for action-conditioned world-model training, but they supervise noisy vision "
+            "tokens rather than preds_action. The remaining work is a pipeline-loaded packer "
+            "check and one-sample forward-dynamics overfit; action-token prediction needs a "
+            "separate policy or inverse-dynamics target export."
+        ),
+    }
+def cosmos3_super_packer_entry() -> dict[str, Any] | None:
+    paths = sorted(
+        (ROOT / "results/omni_finetune").glob("xperience10m_cosmos3_super_action_packer_*/packer_summary.json")
+    )
+    if not paths:
+        return None
+    payloads = [(path, load_json(path)) for path in paths]
+    path, payload = max(payloads, key=lambda item: item[1].get("finished_at_unix") or 0)
+    row_contract = payload.get("row_contract", {}) if isinstance(payload.get("row_contract"), dict) else {}
+    pack_result = payload.get("pack_result", {}) if isinstance(payload.get("pack_result"), dict) else {}
+    return {
+        "id": payload.get("run_id", path.parent.name),
+        "title": "Cosmos3-Super Action Batch Packer Smoke",
+        "scope_label": "batch packer",
+        "scope": "one selected train row from the camera_pose forward_dynamics augmented JSONL",
+        "status": payload.get("status", "unknown"),
+        "source": rel(path),
+        "split": row_contract.get("split"),
+        "counts": {
+            "samples": 1,
+            "raw_action_rows": (row_contract.get("raw_actions_shape") or [None, None])[0],
+            "raw_action_dim": row_contract.get("raw_action_dim"),
+        },
+        "primary_metrics": {
+            "mode": row_contract.get("mode"),
+            "loss_surface": row_contract.get("loss_surface"),
+            "pipeline_loaded": pack_result.get("pipeline_loaded"),
+            "weights_updated": payload.get("weights_updated"),
+        },
+        "weights": "none; schema-only packer smoke, no adapter checkpoint",
+        "interpretation": (
+            "The selected row maps to a camera_pose forward_dynamics contract. In the installed Cosmos3 pipeline this "
+            "uses raw actions as conditioning and supervises noisy vision tokens; it does not supervise preds_action."
         ),
     }
     cosmos_nano_branches = [branch for branch in branches if branch.get("backbone") == "cosmos_world_model"]
     cosmos_super_branches = [branch for branch in branches if branch.get("backbone") == "cosmos3_super_reasoner"]
     cosmos_super_readiness = cosmos3_super_readiness_entry()
+    cosmos_super_action_contract = cosmos3_super_action_contract_entry()
+    cosmos_super_packer = cosmos3_super_packer_entry()
     if qwen_branches:
         current_qwen = max(qwen_branches, key=lambda item: item.get("primary_metrics", {}).get("json_validity_rate") or -1)
         for branch in qwen_branches:
                     ),
                 }
             ],
+            "readiness_runs": [
+                entry for entry in (cosmos_super_readiness, cosmos_super_action_contract, cosmos_super_packer) if entry
+            ],
             "multi_episode_128_runs": cosmos_super_branches,
             "comparison_note": (
                 "Cosmos3-Super is now represented by a verified 448-window held-out "
                 "Reasoner evaluation on the same JSON task as Qwen3. It uses staged base "
                 "weights through vLLM, so it is a model-branch diagnostic, not a weight release. "
+                "A camera-pose proxy forward-dynamics target export now passes the contract audit "
+                "and schema-only packer smoke; true Cosmos3-Super fine-tuning is still not launched "
+                "until the pipeline-loaded packer check and one-sample overfit exist."
             ),
         },
     ]
         "version_reading_notes": [
             "Version 1 is the public-sample 12-task harness with minimal and neural heads.",
             "Version 2 is the selected 128-episode same-split simple/NN baseline alignment.",
+            "Version 3 is the verified model-branch layer: the current final Qwen3-Omni LoRA package is the JSON-task diagnostic result, Cosmos3-Nano is a future-window compatibility result, and Cosmos3-Super Reasoner is a base-weight JSON-task evaluation; Cosmos3-Super now has a camera-pose forward-dynamics contract audit and schema-only packer smoke, but no new fine-tuned weight release.",
         ],
         "versions": versions,
         "model_groups": model_groups,
             "Task-head baselines have both a one-episode public-sample run and a 128-episode same-split metadata/text run.",
             "Qwen3-Omni has a one-episode sensor-adapter smoke test and separate 128-episode LoRA diagnostic packages; only the final 128-episode adapter belongs in the Qwen LoRA model repo.",
             "Cosmos3-Nano has a 128-episode future-window compatibility package.",
+            "Cosmos3-Super has a 128-episode base-weight Reasoner evaluation on the JSON task plus a camera-pose forward-dynamics contract audit; create a separate Cosmos model repo only after real Cosmos adapter/fine-tuned weights exist.",
         ],
         "pending": [
             "Use the final Qwen3 full-eval package as the current Qwen result; older Qwen package rows remain historical diagnostics for comparison.",
+            "Promote Cosmos3 from Nano compatibility, Super base-weight evaluation, and the camera-pose forward-dynamics contract to true fine-tuning only after the pipeline-loaded packer check and one-sample overfit produce new weights.",
         ],
     }
     pieces = []
     for label, keys in (
         ("episodes", ("episodes", "dataset_episodes", "held_out_episode_count")),
+        ("windows/samples", ("windows", "rows", "dataset_samples", "eval_samples", "samples")),
         ("eval", ("eval_samples",)),
     ):
         value = next((counts.get(key) for key in keys if counts.get(key) is not None), None)
         "contact_accuracy",
         "accuracy",
         "macro_f1",
+        "domain_name",
+        "raw_action_dim",
+        "mode",
+        "valid_action_targets",
+        "loss_surface",
+        "pipeline_loaded",
         "diffusers_runtime_supported",
         "chat_sft_supported",
         "weights_updated",
     for entry in group.get("one_episode_runs", []):
         rows.append(("1 episode", entry))
     for entry in group.get("readiness_runs", []):
+        rows.append((entry.get("scope_label", "readiness"), entry))
     for entry in group.get("multi_episode_128_runs", []):
         rows.append(("128 episode", entry))
     for scope, entry in rows:

scripts/omni/export_cosmos3_camera_pose_targets.py ADDED Viewed

	@@ -0,0 +1,250 @@

+#!/usr/bin/env python3
+"""Augment exported Xperience windows with Cosmos3 camera-pose action targets.
+This does not invent robot-control labels. It converts frame-aligned SLAM poses
+from `annotation.hdf5` into the Cosmos3-supported `camera_pose` action domain:
+9D per-transition vectors with translation delta, rotation delta as a rotation
+vector, and absolute displacement from the window start. The target is a
+continuous egocentric-motion proxy suitable for a first Cosmos3 action-packer
+smoke run; it is intentionally separate from the semantic JSON QA target.
+"""
+from __future__ import annotations
+import argparse
+import json
+import math
+from collections import Counter
+from pathlib import Path
+from typing import Any
+import h5py
+import numpy as np
+from qwen3_omni_dataset_utils import load_jsonl, write_jsonl
+RAW_ACTION_DIM = 9
+DOMAIN_NAME = "camera_pose"
+def parse_args() -> argparse.Namespace:
+    workspace_default = Path(__file__).resolve().parents[2]
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--dataset-jsonl", type=Path, required=True)
+    parser.add_argument("--output-jsonl", type=Path, required=True)
+    parser.add_argument("--output-manifest", type=Path, required=True)
+    parser.add_argument("--chunk-size", type=int, default=8)
+    parser.add_argument("--resolution-tier", type=int, default=480, choices=[256, 480, 704, 720])
+    parser.add_argument("--view-point", default="ego_view")
+    parser.add_argument("--max-records", type=int, default=0)
+    parser.add_argument("--strict", action="store_true")
+    return parser.parse_args()
+def read_pose_cache(annotation_path: Path) -> dict[str, np.ndarray]:
+    with h5py.File(annotation_path, "r") as h5:
+        slam = h5["slam"]
+        trans = np.asarray(slam["trans_xyz"], dtype=np.float64)
+        quat = np.asarray(slam["quat_wxyz"], dtype=np.float64)
+        frame_numbers = np.asarray(h5["video"]["frame_number"], dtype=np.int64)
+    return {"trans": trans, "quat": normalize_quat_array(quat), "frame_numbers": frame_numbers}
+def normalize_quat_array(quat: np.ndarray) -> np.ndarray:
+    norm = np.linalg.norm(quat, axis=-1, keepdims=True)
+    norm[norm <= 1e-12] = 1.0
+    quat = quat / norm
+    # Keep quaternion sign continuous enough for simple deltas.
+    for idx in range(1, len(quat)):
+        if np.dot(quat[idx - 1], quat[idx]) < 0:
+            quat[idx] *= -1.0
+    return quat
+def quat_inverse(q: np.ndarray) -> np.ndarray:
+    return np.asarray([q[0], -q[1], -q[2], -q[3]], dtype=np.float64) / max(float(np.dot(q, q)), 1e-12)
+def quat_multiply(a: np.ndarray, b: np.ndarray) -> np.ndarray:
+    aw, ax, ay, az = a
+    bw, bx, by, bz = b
+    return np.asarray(
+        [
+            aw * bw - ax * bx - ay * by - az * bz,
+            aw * bx + ax * bw + ay * bz - az * by,
+            aw * by - ax * bz + ay * bw + az * bx,
+            aw * bz + ax * by - ay * bx + az * bw,
+        ],
+        dtype=np.float64,
+    )
+def quat_to_rotvec(q: np.ndarray) -> np.ndarray:
+    q = q / max(float(np.linalg.norm(q)), 1e-12)
+    if q[0] < 0:
+        q = -q
+    w = float(np.clip(q[0], -1.0, 1.0))
+    xyz = q[1:]
+    sin_half = float(np.linalg.norm(xyz))
+    if sin_half < 1e-8:
+        return 2.0 * xyz
+    angle = 2.0 * math.atan2(sin_half, w)
+    if angle > math.pi:
+        angle -= 2.0 * math.pi
+    return xyz / sin_half * angle
+def nearest_index(frame_numbers: np.ndarray, frame: int) -> int:
+    if frame <= int(frame_numbers[0]):
+        return 0
+    if frame >= int(frame_numbers[-1]):
+        return len(frame_numbers) - 1
+    return int(np.searchsorted(frame_numbers, frame, side="left"))
+def sampled_frame_pairs(start_frame: int, end_frame: int, chunk_size: int) -> list[tuple[int, int]]:
+    if chunk_size < 1:
+        raise ValueError("chunk_size must be >= 1")
+    if end_frame <= start_frame:
+        end_frame = start_frame + chunk_size
+    points = np.linspace(start_frame, end_frame, chunk_size + 1)
+    frames = [int(round(value)) for value in points]
+    pairs: list[tuple[int, int]] = []
+    for left, right in zip(frames[:-1], frames[1:]):
+        if right <= left:
+            right = left + 1
+        pairs.append((left, right))
+    return pairs
+def camera_pose_actions(pose: dict[str, np.ndarray], start_frame: int, end_frame: int, chunk_size: int) -> list[list[float]]:
+    trans = pose["trans"]
+    quat = pose["quat"]
+    frame_numbers = pose["frame_numbers"]
+    start_idx = nearest_index(frame_numbers, start_frame)
+    origin = trans[start_idx]
+    rows: list[list[float]] = []
+    for left_frame, right_frame in sampled_frame_pairs(start_frame, end_frame, chunk_size):
+        li = nearest_index(frame_numbers, left_frame)
+        ri = nearest_index(frame_numbers, right_frame)
+        delta_t = trans[ri] - trans[li]
+        delta_q = quat_multiply(quat[ri], quat_inverse(quat[li]))
+        delta_r = quat_to_rotvec(delta_q)
+        displacement = trans[ri] - origin
+        row = np.concatenate([delta_t, delta_r, displacement]).astype(np.float32)
+        if row.shape[0] != RAW_ACTION_DIM:
+            raise AssertionError(row.shape)
+        rows.append([float(value) for value in row])
+    return rows
+def media_condition(row: dict[str, Any]) -> dict[str, Any]:
+    media = row.get("media") if isinstance(row.get("media"), dict) else {}
+    return {
+        "mosaic_video_path": media.get("mosaic_video_path"),
+        "video_paths": media.get("video_paths") if isinstance(media.get("video_paths"), list) else [],
+        "context_start_frame": media.get("context_start_frame"),
+        "context_end_frame": media.get("context_end_frame"),
+    }
+def augment_rows(rows: list[dict[str, Any]], args: argparse.Namespace) -> tuple[list[dict[str, Any]], dict[str, Any]]:
+    pose_cache: dict[str, dict[str, np.ndarray]] = {}
+    counters = Counter()
+    issues: list[dict[str, Any]] = []
+    augmented: list[dict[str, Any]] = []
+    selected = rows[: args.max_records] if args.max_records > 0 else rows
+    for idx, row in enumerate(selected):
+        counters["rows_seen"] += 1
+        episode_path_raw = row.get("episode_path")
+        window = row.get("center_window") if isinstance(row.get("center_window"), dict) else {}
+        if not episode_path_raw or "start_frame" not in window or "end_frame" not in window:
+            counters["rows_skipped_missing_source_fields"] += 1
+            issues.append({"row_index": idx, "id": row.get("id"), "reason": "missing episode_path or center_window"})
+            if args.strict:
+                raise ValueError(issues[-1])
+            continue
+        annotation_path = Path(str(episode_path_raw)) / "annotation.hdf5"
+        if not annotation_path.exists():
+            counters["rows_skipped_missing_annotation"] += 1
+            issues.append({"row_index": idx, "id": row.get("id"), "reason": f"missing {annotation_path}"})
+            if args.strict:
+                raise FileNotFoundError(annotation_path)
+            continue
+        key = str(annotation_path)
+        if key not in pose_cache:
+            pose_cache[key] = read_pose_cache(annotation_path)
+        start_frame = int(window["start_frame"])
+        end_frame = int(window["end_frame"])
+        try:
+            raw_actions = camera_pose_actions(pose_cache[key], start_frame, end_frame, args.chunk_size)
+        except Exception as exc:
+            counters["rows_skipped_action_build_error"] += 1
+            issues.append({"row_index": idx, "id": row.get("id"), "reason": repr(exc)})
+            if args.strict:
+                raise
+            continue
+        copied = dict(row)
+        copied["cosmos_action_target"] = {
+            "mode": "forward_dynamics",
+            "domain_name": DOMAIN_NAME,
+            "chunk_size": args.chunk_size,
+            "raw_action_dim": RAW_ACTION_DIM,
+            "raw_actions": raw_actions,
+            "resolution_tier": args.resolution_tier,
+            "view_point": args.view_point,
+            "source": {
+                "kind": "slam_camera_pose_delta_proxy_v1",
+                "annotation_hdf5": str(annotation_path),
+                "frame_range": {"start_frame": start_frame, "end_frame": end_frame},
+                "fields": [
+                    "slam/trans_xyz delta",
+                    "slam/quat_wxyz delta as rotation vector",
+                    "slam/trans_xyz displacement from window start",
+                ],
+                "units": "translation in annotation coordinate units; rotation in radians",
+            },
+            "conditioning": media_condition(row),
+        }
+        augmented.append(copied)
+        counters["rows_augmented"] += 1
+    manifest = {
+        "status": "pass" if counters["rows_augmented"] else "fail",
+        "input_dataset_jsonl": str(args.dataset_jsonl),
+        "output_jsonl": str(args.output_jsonl),
+        "domain_name": DOMAIN_NAME,
+        "raw_action_dim": RAW_ACTION_DIM,
+        "chunk_size": args.chunk_size,
+        "resolution_tier": args.resolution_tier,
+        "view_point": args.view_point,
+        "target_kind": "slam_camera_pose_delta_proxy_v1",
+        "counts": dict(counters),
+        "episode_annotation_files_read": len(pose_cache),
+        "issues": issues[:100],
+        "limitations": [
+            "This is an egocentric camera-motion proxy, not a robot gripper or human hand-control action.",
+            "Use it for Cosmos3 action-packer and one-episode overfit smoke tests before claiming model-quality improvement.",
+            "Fit any normalization on train episodes only before a full publishable Cosmos adapter run.",
+        ],
+    }
+    return augmented, manifest
+def main() -> int:
+    args = parse_args()
+    rows = load_jsonl(args.dataset_jsonl)
+    augmented, manifest = augment_rows(rows, args)
+    args.output_jsonl.parent.mkdir(parents=True, exist_ok=True)
+    args.output_manifest.parent.mkdir(parents=True, exist_ok=True)
+    write_jsonl(args.output_jsonl, augmented)
+    args.output_manifest.write_text(json.dumps(manifest, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+    print(json.dumps(manifest, indent=2, ensure_ascii=False))
+    return 0 if manifest["status"] == "pass" else 1
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/pack_cosmos3_super_action_batch.py ADDED Viewed

	@@ -0,0 +1,459 @@

+#!/usr/bin/env python3
+"""Pack one Cosmos3-Super action-conditioning batch from Xperience windows.
+This is the bridge between the public-safe Xperience JSONL export and a real
+Cosmos3 Diffusers trainer. It can run in two modes:
+- schema mode: validate the selected row and infer the supervised loss surface
+  without loading the huge model.
+- pipeline mode: load Cosmos3OmniPipeline and call the installed
+  prepare_latents/_prepare_*_segment helpers to verify tensor shapes and loss
+  indexes for one sample.
+The current camera_pose target export uses mode=forward_dynamics. In the
+installed Cosmos3 pipeline that mode treats actions as conditioning and
+supervises noisy vision tokens, not preds_action. Policy/inverse-dynamics action
+prediction requires a separate target export mode.
+"""
+from __future__ import annotations
+import argparse
+import json
+import time
+from pathlib import Path
+from typing import Any
+from qwen3_omni_dataset_utils import load_jsonl
+ACTION_TARGET_KEYS = (
+    "cosmos_action_target",
+    "cosmos3_action_target",
+    "cosmos_action_condition",
+    "action_target",
+)
+def parse_args() -> argparse.Namespace:
+    workspace_default = Path(__file__).resolve().parents[2]
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--workspace", type=Path, default=workspace_default)
+    parser.add_argument("--dataset-jsonl", type=Path, required=True)
+    parser.add_argument("--run-id", default="xperience10m_cosmos3_super_action_packer_smoke")
+    parser.add_argument("--output-dir", type=Path)
+    parser.add_argument("--model-dir", type=Path)
+    parser.add_argument(
+        "--backbone-config",
+        type=Path,
+        default=workspace_default / "configs" / "omni_backbones" / "cosmos3_super_reasoner.json",
+    )
+    parser.add_argument("--split", default="train")
+    parser.add_argument("--sample-index", type=int, default=0)
+    parser.add_argument("--sample-id")
+    parser.add_argument("--prompt", default="Predict the embodied future under the provided camera-pose action condition.")
+    parser.add_argument("--negative-prompt")
+    parser.add_argument("--fps", type=float, default=24.0)
+    parser.add_argument("--device", default="cuda")
+    parser.add_argument("--dtype", default="bfloat16", choices=["bfloat16", "float16", "float32"])
+    parser.add_argument("--load-pipeline", action="store_true")
+    parser.add_argument("--local-files-only", action=argparse.BooleanOptionalAction, default=True)
+    parser.add_argument("--require-media-exists", action="store_true")
+    return parser.parse_args()
+def dtype_from_name(name: str):
+    import torch
+    return {
+        "bfloat16": torch.bfloat16,
+        "float16": torch.float16,
+        "float32": torch.float32,
+    }[name]
+def write_json(path: Path, payload: dict[str, Any]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(json.dumps(payload, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
+def append_jsonl(path: Path, payload: dict[str, Any]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("a", encoding="utf-8") as handle:
+        handle.write(json.dumps(payload, sort_keys=True, ensure_ascii=False) + "\n")
+def read_json(path: Path) -> dict[str, Any]:
+    if not path.exists():
+        return {}
+    return json.loads(path.read_text(encoding="utf-8"))
+def find_action_target(row: dict[str, Any]) -> tuple[str | None, dict[str, Any] | None]:
+    for key in ACTION_TARGET_KEYS:
+        value = row.get(key)
+        if isinstance(value, dict):
+            return key, value
+    return None, None
+def selected_row(rows: list[dict[str, Any]], args: argparse.Namespace) -> dict[str, Any]:
+    candidates = [row for row in rows if row.get("split") == args.split and find_action_target(row)[1] is not None]
+    if args.sample_id:
+        for row in rows:
+            if row.get("id") == args.sample_id:
+                return row
+        raise ValueError(f"sample id not found: {args.sample_id}")
+    if not candidates:
+        raise ValueError(f"no rows with action targets found for split={args.split!r}")
+    if args.sample_index < 0 or args.sample_index >= len(candidates):
+        raise ValueError(f"sample-index {args.sample_index} outside 0..{len(candidates)-1}")
+    return candidates[args.sample_index]
+def numeric_matrix(value: Any) -> tuple[bool, tuple[int, int] | None]:
+    if not isinstance(value, list) or not value:
+        return False, None
+    width = None
+    for item in value:
+        if not isinstance(item, list) or not item:
+            return False, None
+        width = len(item) if width is None else width
+        if len(item) != width:
+            return False, None
+        for number in item:
+            if not isinstance(number, (int, float)):
+                return False, None
+    return True, (len(value), int(width or 0))
+def media_video_path(row: dict[str, Any], target: dict[str, Any]) -> str | None:
+    conditioning = target.get("conditioning") if isinstance(target.get("conditioning"), dict) else {}
+    media = row.get("media") if isinstance(row.get("media"), dict) else {}
+    for block in (conditioning, media):
+        value = block.get("mosaic_video_path")
+        if value:
+            return str(value)
+    for block in (conditioning, media):
+        paths = block.get("video_paths")
+        if isinstance(paths, list):
+            for item in paths:
+                if isinstance(item, dict) and item.get("path"):
+                    return str(item["path"])
+    return None
+def row_contract(row: dict[str, Any], require_media_exists: bool) -> dict[str, Any]:
+    key, target = find_action_target(row)
+    if target is None:
+        raise ValueError(f"row has no Cosmos action target: {row.get('id')}")
+    video_path = media_video_path(row, target)
+    if not video_path:
+        raise ValueError(f"row has no video conditioning path: {row.get('id')}")
+    if require_media_exists and not Path(video_path).exists():
+        raise FileNotFoundError(video_path)
+    mode = str(target.get("mode"))
+    domain_name = str(target.get("domain_name"))
+    chunk_size = int(target.get("chunk_size"))
+    raw_actions = target.get("raw_actions")
+    ok, shape = numeric_matrix(raw_actions)
+    raw_action_dim = int(target.get("raw_action_dim") or (shape[1] if shape else 0))
+    issues: list[str] = []
+    if mode not in {"forward_dynamics", "policy", "inverse_dynamics"}:
+        issues.append(f"unsupported mode={mode!r}")
+    if domain_name != "camera_pose":
+        issues.append(f"expected camera_pose target for this export, got {domain_name!r}")
+    if chunk_size < 1:
+        issues.append("chunk_size must be >= 1")
+    if mode == "forward_dynamics":
+        if not ok:
+            issues.append("forward_dynamics requires numeric raw_actions")
+        elif shape and shape[1] != raw_action_dim:
+            issues.append(f"raw_actions width {shape[1]} does not match raw_action_dim {raw_action_dim}")
+    if mode == "forward_dynamics":
+        loss_surface = "vision_velocity_conditioned_on_camera_pose"
+        action_loss_expected = False
+        note = (
+            "Cosmos3 forward_dynamics consumes raw_actions as conditioning and predicts noisy vision tokens. "
+            "It does not supervise preds_action for this target mode."
+        )
+    else:
+        loss_surface = "action_velocity"
+        action_loss_expected = True
+        note = (
+            "Cosmos3 policy/inverse_dynamics can expose noisy action tokens, but the current camera-pose export "
+            "does not yet create that target mode."
+        )
+    return {
+        "row_id": row.get("id"),
+        "episode_id": row.get("episode_id"),
+        "split": row.get("split"),
+        "target_key": key,
+        "mode": mode,
+        "domain_name": domain_name,
+        "chunk_size": chunk_size,
+        "raw_action_dim": raw_action_dim,
+        "raw_actions_shape": list(shape) if shape else None,
+        "video_path": video_path,
+        "video_path_exists": Path(video_path).exists(),
+        "loss_surface": loss_surface,
+        "action_loss_expected": action_loss_expected,
+        "interpretation": note,
+        "issues": issues,
+    }
+def instantiate_action_condition(row: dict[str, Any], contract: dict[str, Any]):
+    import torch
+    from diffusers.pipelines.cosmos.pipeline_cosmos3_omni import CosmosActionCondition
+    _, target = find_action_target(row)
+    if target is None:
+        raise ValueError("missing action target")
+    raw_actions = None
+    if target.get("raw_actions") is not None:
+        raw_actions = torch.tensor(target["raw_actions"], dtype=torch.float32)
+    video = [contract["video_path"]]
+    return CosmosActionCondition(
+        mode=contract["mode"],
+        chunk_size=int(contract["chunk_size"]),
+        domain_name=contract["domain_name"],
+        resolution_tier=int(target.get("resolution_tier", 480)),
+        raw_actions=raw_actions,
+        video=video,
+        view_point=str(target.get("view_point", "ego_view")),
+    )
+def resolve_action_canvas(pipe, action) -> tuple[int | None, int | None]:
+    try:
+        from diffusers.pipelines.cosmos.pipeline_cosmos3_omni import _ACTION_RESOLUTION_BINS, VideoProcessor
+        conditioning_clip = [action.image] if action.image is not None else action.video
+        probe = pipe.video_processor.preprocess_video(conditioning_clip)
+        source_h, source_w = int(probe.shape[-2]), int(probe.shape[-1])
+        resolution_key = str(action.resolution_tier)
+        return VideoProcessor.classify_height_width_bin(source_h, source_w, ratios=_ACTION_RESOLUTION_BINS[resolution_key])
+    except Exception:
+        return None, None
+def tokenize_prompt(pipe, args: argparse.Namespace, action, height: int | None, width: int | None) -> list[int]:
+    if hasattr(pipe, "tokenize_prompt"):
+        cond_ids, _ = pipe.tokenize_prompt(
+            args.prompt,
+            args.negative_prompt,
+            num_frames=action.chunk_size + 1,
+            height=height,
+            width=width,
+            fps=args.fps,
+            action_mode=action.mode,
+            action_view_point=action.view_point,
+        )
+        return list(cond_ids)
+    encoded = pipe.tokenizer(args.prompt, add_special_tokens=True)
+    return list(encoded["input_ids"])
+def pack_with_pipeline(row: dict[str, Any], contract: dict[str, Any], args: argparse.Namespace) -> dict[str, Any]:
+    import torch
+    from diffusers import Cosmos3OmniPipeline
+    if args.model_dir is None:
+        raise ValueError("--model-dir is required with --load-pipeline")
+    dtype = dtype_from_name(args.dtype)
+    pipe = Cosmos3OmniPipeline.from_pretrained(
+        str(args.model_dir),
+        torch_dtype=dtype,
+        local_files_only=args.local_files_only,
+    )
+    pipe.to(args.device)
+    if hasattr(pipe, "set_progress_bar_config"):
+        pipe.set_progress_bar_config(disable=True)
+    action = instantiate_action_condition(row, contract)
+    height, width = resolve_action_canvas(pipe, action)
+    input_ids = tokenize_prompt(pipe, args, action, height, width)
+    text_segment = pipe._prepare_text_segment(input_ids, device=args.device)
+    (
+        latents,
+        sound_latents,
+        action_latents,
+        fps_vision,
+        fps_sound,
+        vision_condition_mask,
+        sound_condition_mask,
+        action_condition_mask,
+        action_domain_id,
+        action_image_size,
+        raw_action_dim_resolved,
+        action_condition_frame_indexes,
+    ) = pipe.prepare_latents(
+        num_frames=action.chunk_size + 1,
+        height=height,
+        width=width,
+        fps=args.fps,
+        device=args.device,
+        dtype=dtype,
+        enable_sound=False,
+        action=action,
+    )
+    vision_condition_indexes = torch.nonzero(vision_condition_mask[:, 0, 0] > 0, as_tuple=False).flatten()
+    vision_condition_indexes = [int(idx.item()) for idx in vision_condition_indexes]
+    vision_segment = pipe._prepare_vision_segment(
+        input_vision_tokens=latents,
+        has_image_condition=bool(vision_condition_indexes),
+        mrope_offset=text_segment["vision_start_temporal_offset"],
+        vision_fps=fps_vision,
+        curr=text_segment["und_len"],
+        device=args.device,
+        condition_frame_indexes=vision_condition_indexes,
+    )
+    action_segment = {}
+    if action_latents is not None:
+        action_segment = pipe._prepare_action_segment(
+            input_action_tokens=action_latents,
+            condition_frame_indexes=action_condition_frame_indexes,
+            mrope_offset=text_segment["vision_start_temporal_offset"],
+            action_fps=fps_vision,
+            curr=text_segment["und_len"] + vision_segment["num_vision_tokens"],
+            device=args.device,
+        )
+    action_loss_tokens = int(action_segment.get("action_mse_loss_indexes", torch.tensor([])).numel())
+    vision_loss_tokens = int(vision_segment.get("vision_mse_loss_indexes", torch.tensor([])).numel())
+    status = "pass"
+    if contract["mode"] == "forward_dynamics" and action_loss_tokens != 0:
+        status = "warning_unexpected_action_loss_tokens"
+    elif contract["mode"] != "forward_dynamics" and action_loss_tokens == 0:
+        status = "warning_no_action_loss_tokens"
+    return {
+        "status": status,
+        "pipeline_loaded": True,
+        "model_dir": str(args.model_dir),
+        "dtype": args.dtype,
+        "device": args.device,
+        "canvas": {"height": height, "width": width},
+        "text_tokens": int(text_segment["und_len"]),
+        "vision_latents_shape": list(latents.shape),
+        "vision_condition_frames": vision_condition_indexes,
+        "vision_loss_tokens": vision_loss_tokens,
+        "action_latents_shape": list(action_latents.shape) if action_latents is not None else None,
+        "action_condition_frames": list(action_condition_frame_indexes),
+        "action_loss_tokens": action_loss_tokens,
+        "raw_action_dim_resolved": raw_action_dim_resolved,
+        "action_domain_id": action_domain_id.detach().cpu().tolist() if action_domain_id is not None else None,
+        "loss_surface": contract["loss_surface"],
+        "training_readout": (
+            "Use a vision velocity/rectified-flow loss for this forward_dynamics camera_pose target."
+            if contract["mode"] == "forward_dynamics"
+            else "Use an action velocity loss for policy/inverse_dynamics targets."
+        ),
+        "unused_optional": {
+            "sound_latents": sound_latents is not None,
+            "fps_sound": fps_sound,
+            "sound_condition_mask": sound_condition_mask is not None,
+            "action_image_size": list(action_image_size.shape) if hasattr(action_image_size, "shape") else None,
+        },
+    }
+def write_report(path: Path, payload: dict[str, Any]) -> None:
+    contract = payload["row_contract"]
+    pack = payload["pack_result"]
+    lines = [
+        "# Cosmos3-Super Action Batch Packer",
+        "",
+        f"- Run id: `{payload['run_id']}`",
+        f"- Row: `{contract.get('row_id')}`",
+        f"- Mode: `{contract.get('mode')}`",
+        f"- Domain: `{contract.get('domain_name')}`",
+        f"- Raw action shape: `{contract.get('raw_actions_shape')}`",
+        f"- Pipeline loaded: `{pack.get('pipeline_loaded')}`",
+        f"- Status: `{payload['status']}`",
+        "",
+        "## Loss Surface",
+        "",
+        f"- `{contract.get('loss_surface')}`",
+        f"- {contract.get('interpretation')}",
+        "",
+        "## Next Step",
+        "",
+    ]
+    if contract.get("mode") == "forward_dynamics":
+        lines.append("- Implement the one-sample overfit with a vision velocity/rectified-flow loss under camera-pose action conditioning.")
+        lines.append("- Add a separate policy or inverse-dynamics target export before claiming supervised action-token prediction.")
+    else:
+        lines.append("- Implement the one-sample overfit with action velocity loss over noisy action tokens.")
+    path.write_text("\n".join(lines) + "\n", encoding="utf-8")
+def main() -> int:
+    args = parse_args()
+    args.workspace = args.workspace.expanduser().resolve()
+    args.dataset_jsonl = args.dataset_jsonl.expanduser().resolve()
+    if args.model_dir is not None:
+        args.model_dir = args.model_dir.expanduser().resolve()
+    output_dir = args.output_dir or args.workspace / "results" / "omni_finetune" / args.run_id
+    output_dir = output_dir.expanduser().resolve()
+    progress_path = output_dir / "progress.jsonl"
+    if progress_path.exists():
+        progress_path.unlink()
+    started = time.time()
+    append_jsonl(progress_path, {"event": "start", "time": started, "run_id": args.run_id})
+    rows = load_jsonl(args.dataset_jsonl)
+    row = selected_row(rows, args)
+    contract = row_contract(row, require_media_exists=args.require_media_exists)
+    append_jsonl(progress_path, {"event": "row_selected", "time": time.time(), "row_id": contract["row_id"]})
+    if contract["issues"]:
+        pack_result = {"status": "blocked_row_contract", "pipeline_loaded": False, "issues": contract["issues"]}
+    elif args.load_pipeline:
+        pack_result = pack_with_pipeline(row, contract, args)
+    else:
+        pack_result = {
+            "status": "schema_ready_pipeline_not_loaded",
+            "pipeline_loaded": False,
+            "loss_surface": contract["loss_surface"],
+            "action_loss_expected": contract["action_loss_expected"],
+        }
+    status = "pass" if not contract["issues"] and not str(pack_result["status"]).startswith("warning") else pack_result["status"]
+    payload = {
+        "run_id": args.run_id,
+        "run_kind": "cosmos3_super_action_batch_packer",
+        "started_at_unix": started,
+        "finished_at_unix": time.time(),
+        "elapsed_seconds": time.time() - started,
+        "dataset_jsonl": str(args.dataset_jsonl),
+        "backbone_config": str(args.backbone_config),
+        "backbone": read_json(args.backbone_config),
+        "status": status,
+        "row_contract": contract,
+        "pack_result": pack_result,
+        "weights_updated": False,
+    }
+    write_json(output_dir / "packer_summary.json", payload)
+    write_json(
+        output_dir / "training_metadata.json",
+        {
+            "run_id": args.run_id,
+            "run_kind": payload["run_kind"],
+            "weights_updated": False,
+            "checkpoint_dir": None,
+            "status": status,
+            "loss_surface": contract["loss_surface"],
+        },
+    )
+    write_report(output_dir / "RUN_REPORT.md", payload)
+    append_jsonl(progress_path, {"event": "complete", "time": time.time(), "status": status})
+    print(json.dumps({"status": status, "output_dir": str(output_dir)}, indent=2))
+    return 0 if status == "pass" else 1
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/run_qwen3_omni_v4_4epoch_8gpu.sh ADDED Viewed

	@@ -0,0 +1,105 @@

+#!/usr/bin/env bash
+set -euo pipefail
+# Stronger Qwen3-Omni LoRA continuation over the already exported 128-episode
+# 96/16/16 dataset. This launcher intentionally reuses the sealed split and
+# writes a distinct run id so it cannot overwrite the public v3 diagnostic.
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_DIR="${PROJECT_DIR:-$(cd "$SCRIPT_DIR/../.." && pwd)}"
+cd "$PROJECT_DIR"
+RUN_ID="${RUN_ID:-xperience10m_qwen3_omni_128ep_structured_json_v4_4epoch_full8gpu_lora}"
+DATASET_JSONL="${DATASET_JSONL:-results/omni_finetune/xperience10m_qwen3_omni_128ep_96train_16val_16test_valmon_20260605_dataset/dataset.jsonl}"
+MODEL_ID="${MODEL_ID:-$HOME/Ropedia/modelscope_models/Qwen__Qwen3-Omni-30B-A3B-Instruct}"
+BACKBONE_CONFIG="${BACKBONE_CONFIG:-configs/omni_backbones/qwen3_omni_lora.json}"
+EPOCHS="${EPOCHS:-4}"
+GRADIENT_ACCUMULATION_STEPS="${GRADIENT_ACCUMULATION_STEPS:-8}"
+MAX_VAL_SAMPLES="${MAX_VAL_SAMPLES:-512}"
+RUN_DIR="results/omni_finetune/${RUN_ID}"
+LOG="${RUN_DIR}/train.launch.log"
+STATUS="${RUN_DIR}/launch_status.jsonl"
+mkdir -p "$RUN_DIR"
+json_status() {
+  .venv/bin/python - "$STATUS" "$@" <<'PY'
+import json
+import sys
+import time
+path = sys.argv[1]
+payload = {"time": time.time()}
+for item in sys.argv[2:]:
+    key, value = item.split("=", 1)
+    if value.isdigit():
+        value = int(value)
+    payload[key] = value
+with open(path, "a", encoding="utf-8") as handle:
+    handle.write(json.dumps(payload, sort_keys=True) + "\n")
+print(json.dumps(payload, sort_keys=True), flush=True)
+PY
+}
+if [[ ! -s "$DATASET_JSONL" ]]; then
+  json_status event=blocked_missing_dataset dataset_jsonl="$DATASET_JSONL"
+  exit 2
+fi
+if pgrep -af "train_qwen3_omni_lora.py.*--run-id ${RUN_ID}" >/dev/null 2>&1; then
+  json_status event=already_running run_id="$RUN_ID"
+  pgrep -af "train_qwen3_omni_lora.py.*--run-id ${RUN_ID}"
+  exit 0
+fi
+if pgrep -af "train_qwen3_omni_lora.py" >/dev/null 2>&1; then
+  json_status event=blocked_other_training run_id="$RUN_ID"
+  pgrep -af "train_qwen3_omni_lora.py"
+  exit 3
+fi
+cmd=(
+  .venv/bin/python -m accelerate.commands.launch
+  --num_processes 8
+  --mixed_precision bf16
+  --use_fsdp
+  --fsdp_sharding_strategy FULL_SHARD
+  --fsdp_auto_wrap_policy TRANSFORMER_BASED_WRAP
+  --fsdp_transformer_layer_cls_to_wrap Qwen3OmniMoeThinkerTextDecoderLayer
+  --fsdp_use_orig_params true
+  --fsdp_cpu_ram_efficient_loading true
+  --fsdp_sync_module_states true
+  --fsdp_activation_checkpointing true
+  scripts/omni/train_qwen3_omni_lora.py
+  --dataset-jsonl "$DATASET_JSONL"
+  --model-id "$MODEL_ID"
+  --backbone-config "$BACKBONE_CONFIG"
+  --run-id "$RUN_ID"
+  --train-split train
+  --val-split val
+  --epochs "$EPOCHS"
+  --batch-size 1
+  --gradient-accumulation-steps "$GRADIENT_ACCUMULATION_STEPS"
+  --max-train-samples 0
+  --max-val-samples "$MAX_VAL_SAMPLES"
+  --local-files-only
+  --gradient-checkpointing
+  --progress-every 10
+)
+json_status event=launch_start run_id="$RUN_ID" epochs="$EPOCHS" dataset_jsonl="$DATASET_JSONL"
+CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0,1,2,3,4,5,6,7}" \
+PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-expandable_segments:True}" \
+nohup "${cmd[@]}" > "$LOG" 2>&1 < /dev/null &
+pid=$!
+sleep 3
+if ps -p "$pid" >/dev/null 2>&1; then
+  json_status event=launch_detached run_id="$RUN_ID" pid="$pid" log="$LOG"
+  echo "launched run_id=${RUN_ID} pid=${pid} log=${LOG}"
+  exit 0
+fi
+json_status event=launch_failed run_id="$RUN_ID" log="$LOG"
+tail -120 "$LOG" || true
+exit 1

scripts/verify_live_publication.py CHANGED Viewed

@@ -311,7 +311,7 @@ MARKER_CHECKS = [
             "100.00%",
             "omni_model_comparison.json",
             "ropedia-qwen3-omni-lora-128ep",
-            "Cosmos3-Super has a verified base-weight Reasoner JSON-task evaluation",
         ],
         "forbidden": [
             "xperience10m-" + "taskfirst-v10",
@@ -340,7 +340,7 @@ MARKER_CHECKS = [
             "100.00%",
             "omni_model_comparison.json",
             "ropedia-qwen3-omni-lora-128ep",
-            "Cosmos3-Super has a verified base-weight Reasoner JSON-task evaluation",
         ],
         "forbidden": [
             "xperience10m-" + "taskfirst-v10",

             "100.00%",
             "omni_model_comparison.json",
             "ropedia-qwen3-omni-lora-128ep",
+            "Cosmos3-Super has a verified base-weight JSON-task evaluation plus a camera-pose forward-dynamics contract audit",
         ],
         "forbidden": [
             "xperience10m-" + "taskfirst-v10",
             "100.00%",
             "omni_model_comparison.json",
             "ropedia-qwen3-omni-lora-128ep",
+            "Cosmos3-Super has a verified base-weight JSON-task evaluation plus a camera-pose forward-dynamics contract audit",
         ],
         "forbidden": [
             "xperience10m-" + "taskfirst-v10",