Add files using upload-large-folder tool

.gitattributes

@@ -66,3 +66,4 @@ results/omni_finetune/a100_128_metadata_task_baselines_20260616_v2/action_object
 results/omni_finetune/a100_128_metadata_task_baselines_20260616_v2/interaction_text_prediction/raw_interaction_label_audit.csv filter=lfs diff=lfs merge=lfs -text
 results/omni_finetune/a100_128_metadata_task_baselines_20260616_v2/interaction_text_prediction/confusion_matrix.csv filter=lfs diff=lfs merge=lfs -text
 results/omni_finetune/a100_128_metadata_task_baselines_20260616_v2/neural_mlp/interaction_text_prediction/confusion_matrix.csv filter=lfs diff=lfs merge=lfs -text
+results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/predictions.jsonl filter=lfs diff=lfs merge=lfs -text

data/artifact_index.json

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-18T22:57:14+00:00",
+  "generated_at_utc": "2026-06-20T14:03:31+00:00",
   "status": "pass",
   "artifact_count": 214,
   "missing": [],
@@ -610,7 +610,7 @@
       "shows": "Machine-readable source-alignment pass/fail check for repo, website, and HF surfaces.",
       "exists": true,
       "bytes": 4432,
-      "sha256": "1df30908392b455cc7263154126d55709cf97c60ff4213f0eee9ce17d92fb234"
+      "sha256": "03e110008eda658d725dc690ababf5e07dab8bab7ec4a6155d237ab9f9f95ecc"
     },
     {
       "id": "source_alignment_validator",
@@ -730,8 +730,8 @@
       "surface": "website_hf",
       "shows": "Stores normalized 20-axis radar values, raw task metrics, Qwen3/Cosmos overlay mappings, branch-card caveats, and explicit scoreless status records.",
       "exists": true,
-      "bytes": 229168,
-      "sha256": "64c7c10f3ce663178f2eea4740bb45de0f94b1246448f14f5521d293be916e77"
+      "bytes": 228639,
+      "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
     },
     {
       "id": "single_episode_task_model_radar_json",
@@ -742,7 +742,7 @@
       "shows": "Machine-readable split radar for the one-episode Minimal and Neural MLP baselines, both scored on all 20 task contracts.",
       "exists": true,
       "bytes": 51064,
-      "sha256": "9d42b045d833e50a2bfa8905203150af1f28ea4736d22edb0fe27e39eedf4574"
+      "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
     },
     {
       "id": "episode128_task_model_radar_json",
@@ -752,8 +752,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable split radar for selected 128-episode metadata/raw baselines and verified Qwen3/Cosmos branches, preserving explicit scoreless cells.",
       "exists": true,
-      "bytes": 185314,
-      "sha256": "5c61c8f9a4a9a7b6bdc65e39ef5fda6117df48412afe260c0be4da501bd7e8db"
+      "bytes": 184785,
+      "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
     },
     {
       "id": "task_method_20_result_matrix_json",
@@ -763,8 +763,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable 9-method by 20-task matrix where every method has 20 records and scoreless cells carry unsupported/not-evaluated reasons.",
       "exists": true,
-      "bytes": 128805,
-      "sha256": "d5572768c6fb671eadf87870d20c8645dfae384f862aa63e929b2db50a5cc51a"
+      "bytes": 128510,
+      "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
     },
     {
       "id": "task_method_20_result_matrix",
@@ -774,8 +774,8 @@
       "surface": "repo_hf",
       "shows": "Reader-facing table that separates 20 records per method from numeric scored axes, documented raw128 proxy scores, unsupported metadata targets, and model targets not evaluated in verified packages.",
       "exists": true,
-      "bytes": 3876,
-      "sha256": "a9ce23a452ca73b6d08219ffaac1782d5baad11f9e3662138b62cb4cd5ff206f"
+      "bytes": 3771,
+      "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
     },
     {
       "id": "task_method_20_gap_audit_json",
@@ -785,8 +785,8 @@
       "surface": "website_hf",
       "shows": "Machine-readable 180-record gap ledger with numeric scores, scoreless cells, explicit status reasons, and next evidence needed before new scores can be published.",
       "exists": true,
-      "bytes": 28259,
-      "sha256": "860dea3d3ce2f2c2146c97918a5deca05ae315238b584c10e53f74a9d2c26dcd"
+      "bytes": 20037,
+      "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
     },
     {
       "id": "task_method_20_gap_audit",
@@ -796,8 +796,8 @@
       "surface": "repo_hf",
       "shows": "Reader-facing ledger that lists every scoreless method-task cell and the concrete target or model-output evidence required before it can become numeric.",
       "exists": true,
-      "bytes": 8428,
-      "sha256": "b33c7f91565a2603b138eb4ec1c6df32d823225b16563eba279f416b0fe2a964"
+      "bytes": 6398,
+      "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
     },
     {
       "id": "unified_task_model_radar_chart",
@@ -807,8 +807,8 @@
       "surface": "website_hf",
       "shows": "Compares minimal and neural MLP baselines across all 20 tasks, with Qwen3/Cosmos task-aligned model overlays.",
       "exists": true,
-      "bytes": 54936,
-      "sha256": "f29db4099d60c093ebd092bee37a2fc1338bbff17d7e52b8a1b27ae6fc06727a"
+      "bytes": 56167,
+      "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
     },
     {
       "id": "single_episode_task_model_radar_chart",
@@ -829,8 +829,8 @@
       "surface": "website_hf",
       "shows": "Separates the selected 128-episode methods: raw-feature simple/NN as complete 20/20 scored polygons and metadata/Qwen/Cosmos as task-aligned overlays.",
       "exists": true,
-      "bytes": 48923,
-      "sha256": "75c389122a413548bdc26fd05cbbad2a85015ebd222ba1a23571f3ff4f0eddcf"
+      "bytes": 50154,
+      "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
     },
     {
       "id": "unified_task_model_radar_builder",
@@ -840,8 +840,8 @@
       "surface": "repo_hf",
       "shows": "Regenerates the direction-aware radar chart and machine-readable metric overlay JSON.",
       "exists": true,
-      "bytes": 57104,
-      "sha256": "805cfce89529fec73edd3e436c1a76afae46cebe31ba5e441b6dfb5cc142c935"
+      "bytes": 61983,
+      "sha256": "01a709eaa65fce0747e4d05b96bba408090dc31dd73819aba54713dad7ad15cf"
     },
     {
       "id": "task_method_20_gap_audit_builder",
@@ -873,8 +873,8 @@
       "surface": "repo_hf",
       "shows": "Checks whether Qwen3/Cosmos branches have train, validation, and test prediction files before extending model overlays to all 20 task contracts.",
       "exists": true,
-      "bytes": 4977,
-      "sha256": "b9caf6d32bb801d845b57b4450b92c10a4c3f9d14890f6a209e5a52aede92568"
+      "bytes": 4770,
+      "sha256": "ad701c03153c6755f284281640465efde3313f48bc0189942f5637bb19328bfb"
     },
     {
       "id": "model_output_probe_script",
@@ -1116,7 +1116,7 @@
       "shows": "Machine-readable release-check summary for validators, mirrors, and public project surfaces.",
       "exists": true,
       "bytes": 8100,
-      "sha256": "b88b46bd9b16d4afaf4836d0acff208b46559a493cfcffa881f1528fa282292c"
+      "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
     },
     {
       "id": "public_surface_qa",
@@ -1297,7 +1297,7 @@
       "volatile": true,
       "shows": "Confirms public bundles exclude raw data, caches, heavy archives, and credential text.",
       "exists": true,
-      "bytes": 9613,
+      "bytes": 9758,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -1321,7 +1321,7 @@
       "volatile": true,
       "shows": "Confirms prepared GitHub/HF Space/artifact/model mirrors share the same critical data, figure, website HTML, and validator files.",
       "exists": true,
-      "bytes": 1015184,
+      "bytes": 1194785,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -1333,7 +1333,7 @@
       "volatile": true,
       "shows": "Confirms local website links, anchors, JSON data files, and referenced images resolve.",
       "exists": true,
-      "bytes": 20021,
+      "bytes": 20022,
       "hash_policy": "existence_and_size_only"
     },
     {

data/episode128_task_model_radar.json

@@ -1,12 +1,12 @@
 {
   "title": "128-Episode 20-Task Radar",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:24:26+00:00",
+  "generated_at_utc": "2026-06-20T13:58:04+00:00",
   "description": "Selected 128-episode metadata/raw baselines plus verified Qwen3/Cosmos branches. Every method has 20 records; numeric scores appear only where the public artifact produced that task target.",
   "task_count": 20,
   "method_count": 7,
   "method_task_record_count": 140,
-  "scored_method_task_count": 119,
+  "scored_method_task_count": 124,
   "normalization_policy": {
     "higher_is_better": "bounded metrics are plotted directly on 0-1 axes after clipping to [0, 1]",
     "lower_is_better": "lower-error metrics are converted to best_observed_value / raw_value within the same task",
@@ -146,20 +146,20 @@
       "kind": "partial_128_episode_foundation_model_overlay",
       "scope": "128 selected episodes, held-out test",
       "stroke_dasharray": "4 7",
-      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 8/16 and a derived task-20 action-boundary timing probe scored from existing verified JSON.",
+      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 5/8/9/10/11/12/13/14/16/17/18/19/20 probes where public metrics exist.",
       "plotted_as": "colored point overlay",
       "result_record_count": 20,
-      "scored_task_count": 10,
-      "covered_task_count": 10,
+      "scored_task_count": 15,
+      "covered_task_count": 15,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 10,
+      "scoreless_task_count": 5,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 10,
+      "not_evaluated_task_count": 5,
       "status_counts": {
-        "not_evaluated_in_verified_package": 10,
-        "scored": 10
+        "not_evaluated_in_verified_package": 5,
+        "scored": 15
       },
-      "coverage_fraction": 0.5,
+      "coverage_fraction": 0.75,
       "result_record_fraction": 1.0
     },
     {
@@ -620,15 +620,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mpjpe",
-          "source": null,
+          "raw": 0.8915253522315043,
+          "metric_key": "hand_trajectory_forecast_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/hand_trajectory_forecast/metrics.json",
           "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.12097265238372007,
+          "raw_text": "0.8915",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": null,
@@ -984,15 +984,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mrr",
-          "source": null,
+          "raw": 0.6628490677465636,
+          "metric_key": "cross_modal_retrieval_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/cross_modal_retrieval/metrics.json",
           "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.6628490677465636,
+          "raw_text": "0.6628",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": 0.022138720585222767,
@@ -1075,15 +1075,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "r2",
-          "source": null,
+          "raw": 0.9939466801653591,
+          "metric_key": "modality_reconstruction_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/metrics.json",
           "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.9939466801653591,
+          "raw_text": "0.9939",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": 0.0002873382957286892,
@@ -1803,15 +1803,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mae",
-          "source": null,
+          "raw": 0.9896650636969544,
+          "metric_key": "imu_to_hand_pose_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/imu_to_hand_pose/metrics.json",
           "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.04248852414968175,
+          "raw_text": "0.9897",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": null,
@@ -1894,15 +1894,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mrr",
-          "source": null,
+          "raw": 0.9979751961528727,
+          "metric_key": "camera_view_sync_retrieval_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/camera_view_sync_retrieval/metrics.json",
           "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.9979751961528727,
+          "raw_text": "0.9980",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": null,
@@ -2610,17 +2610,17 @@
       "task_label": "Hand Trajectory Forecasting",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mpjpe",
-      "source": null,
+      "raw": 0.8915253522315043,
+      "raw_text": "0.8915",
+      "normalized_score": 0.12097265238372007,
+      "metric_key": "hand_trajectory_forecast_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/hand_trajectory_forecast/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 5,
@@ -3114,17 +3114,17 @@
       "task_label": "Cross-Modal Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.6628490677465636,
+      "raw_text": "0.6628",
+      "normalized_score": 0.6628490677465636,
+      "metric_key": "cross_modal_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/cross_modal_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 9,
@@ -3240,17 +3240,17 @@
       "task_label": "Cross-Modal Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "r2",
-      "source": null,
+      "raw": 0.9939466801653591,
+      "raw_text": "0.9939",
+      "normalized_score": 0.9939466801653591,
+      "metric_key": "modality_reconstruction_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 10,
@@ -4248,17 +4248,17 @@
       "task_label": "IMU-to-Hand Pose Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mae",
-      "source": null,
+      "raw": 0.9896650636969544,
+      "raw_text": "0.9897",
+      "normalized_score": 0.04248852414968175,
+      "metric_key": "imu_to_hand_pose_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/imu_to_hand_pose/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 18,
@@ -4374,17 +4374,17 @@
       "task_label": "Camera-View Synchronization Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.9979751961528727,
+      "raw_text": "0.9980",
+      "normalized_score": 0.9979751961528727,
+      "metric_key": "camera_view_sync_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/camera_view_sync_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 19,

data/public_surface_qa.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Public Project Surface",
   "status": "pass",
-  "generated_at_utc": "2026-06-19T11:30:39+00:00",
+  "generated_at_utc": "2026-06-20T14:03:48+00:00",
   "scope": "Repo README, GitHub Pages HTML, Hugging Face Space card, artifact dataset card, and model card.",
   "checks": [
     {
@@ -18,7 +18,7 @@
         "website_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-18T22:58:01+00:00"
+          "generated_at_utc": "2026-06-20T13:51:41+00:00"
         },
         "rendered_site_check": {
           "exists": true,
@@ -28,27 +28,27 @@
         "task_surface_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-19T11:30:21+00:00"
+          "generated_at_utc": "2026-06-20T13:51:16+00:00"
         },
         "source_alignment": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-19T11:30:21+00:00"
+          "generated_at_utc": "2026-06-20T13:51:16+00:00"
         },
         "scale_up_status": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-19T11:30:24+00:00"
+          "generated_at_utc": "2026-06-20T13:51:18+00:00"
         },
         "publication_package": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-18T22:58:11+00:00"
+          "generated_at_utc": "2026-06-20T13:51:51+00:00"
         },
         "mirror_parity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-18T23:18:33+00:00"
+          "generated_at_utc": "2026-06-20T04:32:57+00:00"
         }
       },
       "failures": {}

data/quality_gates.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Release Checks",
   "status": "pass",
-  "generated_at_utc": "2026-06-18T22:57:13+00:00",
+  "generated_at_utc": "2026-06-20T14:03:29+00:00",
   "rule": "A release is current when the automated reports pass and the live GitHub/Hugging Face mirrors are verified after publishing.",
   "automated_gates": [
     {

data/scope_claims_audit.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-19T11:30:24+00:00",
+  "generated_at_utc": "2026-06-20T14:04:01+00:00",
   "summary": {
     "qwen3_omni_verified_diagnostic_pilot": true,
     "dataset_manifest_num_episodes": 119,
@@ -9,7 +9,7 @@
     "eval_num_samples": 4032,
     "eval_json_validity_rate": 0.9990079365079365,
     "quality_target_met": true,
-    "historical_identifier_count": 1830,
+    "historical_identifier_count": 1842,
     "public_32_episode_status_file_count": 1,
     "failure_count": 0
   },
@@ -84,7 +84,7 @@
     {
       "name": "historical_32ep_identifiers_are_confined_to_readiness_artifacts",
       "status": "pass",
-      "detail": "historical identifiers found in result provenance files=1830",
+      "detail": "historical identifiers found in result provenance files=1842",
       "evidence": [
         "results/omni_finetune/"
       ]
@@ -420,6 +420,6 @@
       "example": "{\"id\": \"xperience-10m-sample:qa:51\", \"episode_id\": \"xperience-10m-sample\", \"split\": \"train\", \"target\": \"episode_qa\", \"prompt_type\": \"json_episode_understanding\", \"center_window\": {\"start_frame\": 1020, \"end_frame\": 1039, \"num_frames\": 20}, \"media\": {\"video_path"
     }
   ],
-  "historical_identifier_total_count": 1830,
+  "historical_identifier_total_count": 1842,
   "failures": []
 }

data/single_episode_task_model_radar.json

@@ -1,7 +1,7 @@
 {
   "title": "Single-Episode 20-Task Radar",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:24:26+00:00",
+  "generated_at_utc": "2026-06-20T13:58:04+00:00",
   "description": "Minimal and Neural MLP baselines on the one public sample episode, both scored on all 20 task contracts.",
   "task_count": 20,
   "method_count": 2,

data/source_alignment_audit.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Source Alignment Note",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:25:30+00:00",
+  "generated_at_utc": "2026-06-20T14:03:48+00:00",
   "alignment_json": "docs/data/xperience10m_dataset_card_alignment.json",
   "alignment_summary": {
     "full_dataset_repo": "ropedia-ai/xperience-10m",

data/task_method_20_gap_audit.json

@@ -1,10 +1,10 @@
 {
-  "generated_at_utc": "2026-06-20T04:24:42+00:00",
+  "generated_at_utc": "2026-06-20T13:58:05+00:00",
   "immediate_actions": [
     {
       "artifact": "docs/data/task_method_20_gap_audit.json",
       "id": "gap_audit",
-      "purpose": "Keep the 21 scoreless cells visible and reproducible."
+      "purpose": "Keep the 16 scoreless cells visible and reproducible."
     },
     {
       "artifact": "scripts/omni/score_model_output_probes.py",
@@ -37,11 +37,11 @@
       "proxy_scored_task_count": 0,
       "result_record_count": 20,
       "scope": "128 selected episodes, held-out test",
-      "scored_task_count": 10,
-      "scoreless_task_count": 10,
+      "scored_task_count": 15,
+      "scoreless_task_count": 5,
       "status_counts": {
-        "not_evaluated_in_verified_package": 10,
-        "scored": 10
+        "not_evaluated_in_verified_package": 5,
+        "scored": 15
       }
     },
     "metadata128_neural_mlp": {
@@ -135,12 +135,12 @@
   },
   "missing_by_method": {
     "cosmos3_nano_future_window": 9,
-    "cosmos3_super_reasoner": 10,
+    "cosmos3_super_reasoner": 5,
     "metadata128_neural_mlp": 1,
     "metadata128_simple": 1
   },
   "missing_by_status": {
-    "not_evaluated_in_verified_package": 19,
+    "not_evaluated_in_verified_package": 14,
     "not_supported_by_metadata_only_package": 1,
     "unsupported_without_required_target": 1
   },
@@ -149,8 +149,7 @@
       "cosmos3_nano_future_window"
     ],
     "05 Hand Trajectory Forecasting": [
-      "cosmos3_nano_future_window",
-      "cosmos3_super_reasoner"
+      "cosmos3_nano_future_window"
     ],
     "07 Object Relevance Prediction": [
       "cosmos3_nano_future_window"
@@ -158,12 +157,6 @@
     "08 Language Grounding": [
       "cosmos3_nano_future_window"
     ],
-    "09 Cross-Modal Retrieval": [
-      "cosmos3_super_reasoner"
-    ],
-    "10 Cross-Modal Reconstruction": [
-      "cosmos3_super_reasoner"
-    ],
     "11 Temporal Order Verification": [
       "cosmos3_nano_future_window",
       "cosmos3_super_reasoner"
@@ -183,12 +176,10 @@
       "cosmos3_super_reasoner"
     ],
     "18 IMU-to-Hand Pose Reconstruction": [
-      "cosmos3_nano_future_window",
-      "cosmos3_super_reasoner"
+      "cosmos3_nano_future_window"
     ],
     "19 Camera-View Synchronization Retrieval": [
       "cosmos3_nano_future_window",
-      "cosmos3_super_reasoner",
       "metadata128_neural_mlp",
       "metadata128_simple"
     ]
@@ -207,19 +198,6 @@
       "task_label": "Procedure Step Recognition",
       "task_number": 2
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "mpjpe",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-      "recommended_next_step": "Generate verified model outputs for this task contract and score them against the held-out labels.",
-      "scope": "multi_episode_128_partial_model_overlay",
-      "series_id": "cosmos3_super_reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "task_id": "hand_trajectory_forecast",
-      "task_label": "Hand Trajectory Forecasting",
-      "task_number": 5
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "mpjpe",
@@ -259,32 +237,6 @@
       "task_label": "Language Grounding",
       "task_number": 8
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "mrr",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-      "recommended_next_step": "Generate verified model outputs for this task contract and score them against the held-out labels.",
-      "scope": "multi_episode_128_partial_model_overlay",
-      "series_id": "cosmos3_super_reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "task_id": "cross_modal_retrieval",
-      "task_label": "Cross-Modal Retrieval",
-      "task_number": 9
-    },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "r2",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-      "recommended_next_step": "Generate verified model outputs for this task contract and score them against the held-out labels.",
-      "scope": "multi_episode_128_partial_model_overlay",
-      "series_id": "cosmos3_super_reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "task_id": "modality_reconstruction",
-      "task_label": "Cross-Modal Reconstruction",
-      "task_number": 10
-    },
     {
       "method": "Cosmos3-Super Reasoner",
       "metric_key": "f1",
@@ -389,19 +341,6 @@
       "task_label": "Future Object-Set Forecasting",
       "task_number": 17
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "mae",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-      "recommended_next_step": "Generate verified model outputs for this task contract and score them against the held-out labels.",
-      "scope": "multi_episode_128_partial_model_overlay",
-      "series_id": "cosmos3_super_reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "task_id": "imu_to_hand_pose",
-      "task_label": "IMU-to-Hand Pose Reconstruction",
-      "task_number": 18
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "mae",
@@ -441,19 +380,6 @@
       "task_label": "Camera-View Synchronization Retrieval",
       "task_number": 19
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "mrr",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-      "recommended_next_step": "Generate verified model outputs for this task contract and score them against the held-out labels.",
-      "scope": "multi_episode_128_partial_model_overlay",
-      "series_id": "cosmos3_super_reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "task_id": "camera_view_sync_retrieval",
-      "task_label": "Camera-View Synchronization Retrieval",
-      "task_number": 19
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "mrr",
@@ -514,8 +440,8 @@
     "method_count": 9,
     "method_task_record_count": 180,
     "proxy_scored_method_task_count": 4,
-    "scored_method_task_count": 159,
-    "scoreless_method_task_count": 21,
+    "scored_method_task_count": 164,
+    "scoreless_method_task_count": 16,
     "task_count": 20
   },
   "source_matrix": "docs/data/task_method_20_result_matrix.json",

data/task_method_20_result_matrix.json

@@ -1,11 +1,11 @@
 {
   "title": "Task Method 20-Result Matrix",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:24:26+00:00",
+  "generated_at_utc": "2026-06-20T13:58:04+00:00",
   "task_count": 20,
   "method_count": 9,
   "method_task_record_count": 180,
-  "scored_method_task_count": 159,
+  "scored_method_task_count": 164,
   "series": [
     {
       "id": "minimal",
@@ -180,20 +180,20 @@
       "kind": "partial_128_episode_foundation_model_overlay",
       "scope": "128 selected episodes, held-out test",
       "stroke_dasharray": "4 7",
-      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 8/16 and a derived task-20 action-boundary timing probe scored from existing verified JSON.",
+      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 5/8/9/10/11/12/13/14/16/17/18/19/20 probes where public metrics exist.",
       "plotted_as": "colored point overlay",
       "result_record_count": 20,
-      "scored_task_count": 10,
-      "covered_task_count": 10,
+      "scored_task_count": 15,
+      "covered_task_count": 15,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 10,
+      "scoreless_task_count": 5,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 10,
+      "not_evaluated_task_count": 5,
       "status_counts": {
-        "not_evaluated_in_verified_package": 10,
-        "scored": 10
+        "not_evaluated_in_verified_package": 5,
+        "scored": 15
       },
-      "coverage_fraction": 0.5,
+      "coverage_fraction": 0.75,
       "result_record_fraction": 1.0
     },
     {
@@ -1002,17 +1002,17 @@
       "task_label": "Hand Trajectory Forecasting",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mpjpe",
-      "source": null,
+      "raw": 0.8915253522315043,
+      "raw_text": "0.8915",
+      "normalized_score": 0.12097265238372007,
+      "metric_key": "hand_trajectory_forecast_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/hand_trajectory_forecast/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 5,
@@ -1650,17 +1650,17 @@
       "task_label": "Cross-Modal Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.6628490677465636,
+      "raw_text": "0.6628",
+      "normalized_score": 0.6628490677465636,
+      "metric_key": "cross_modal_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/cross_modal_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 9,
@@ -1812,17 +1812,17 @@
       "task_label": "Cross-Modal Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "r2",
-      "source": null,
+      "raw": 0.9939466801653591,
+      "raw_text": "0.9939",
+      "normalized_score": 0.9939466801653591,
+      "metric_key": "modality_reconstruction_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 10,
@@ -3108,17 +3108,17 @@
       "task_label": "IMU-to-Hand Pose Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mae",
-      "source": null,
+      "raw": 0.9896650636969544,
+      "raw_text": "0.9897",
+      "normalized_score": 0.04248852414968175,
+      "metric_key": "imu_to_hand_pose_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/imu_to_hand_pose/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 18,
@@ -3270,17 +3270,17 @@
       "task_label": "Camera-View Synchronization Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.9979751961528727,
+      "raw_text": "0.9980",
+      "normalized_score": 0.9979751961528727,
+      "metric_key": "camera_view_sync_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/camera_view_sync_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 19,

data/task_surface_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:25:09+00:00",
+  "generated_at_utc": "2026-06-20T14:04:25+00:00",
   "summary": {
     "task_count": 12,
     "expected_task_count": 12,

data/unified_task_model_radar.json

@@ -1,11 +1,11 @@
 {
   "title": "Unified 20-Task Model Radar",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:24:26+00:00",
+  "generated_at_utc": "2026-06-20T13:58:04+00:00",
   "task_count": 20,
   "method_count": 9,
   "method_task_record_count": 180,
-  "scored_method_task_count": 159,
+  "scored_method_task_count": 164,
   "normalization_policy": {
     "higher_is_better": "bounded metrics are plotted directly on 0-1 axes after clipping to [0, 1]",
     "lower_is_better": "lower-error metrics are converted to best_observed_value / raw_value within the same task",
@@ -189,20 +189,20 @@
       "kind": "partial_128_episode_foundation_model_overlay",
       "scope": "128 selected episodes, held-out test",
       "stroke_dasharray": "4 7",
-      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 8/16 and a derived task-20 action-boundary timing probe scored from existing verified JSON.",
+      "method_detail": "Verified Cosmos3-Super base-weight Reasoner JSON-task evaluation, plus task 5/8/9/10/11/12/13/14/16/17/18/19/20 probes where public metrics exist.",
       "plotted_as": "colored point overlay",
       "result_record_count": 20,
-      "scored_task_count": 10,
-      "covered_task_count": 10,
+      "scored_task_count": 15,
+      "covered_task_count": 15,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 10,
+      "scoreless_task_count": 5,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 10,
+      "not_evaluated_task_count": 5,
       "status_counts": {
-        "not_evaluated_in_verified_package": 10,
-        "scored": 10
+        "not_evaluated_in_verified_package": 5,
+        "scored": 15
       },
-      "coverage_fraction": 0.5,
+      "coverage_fraction": 0.75,
       "result_record_fraction": 1.0
     },
     {
@@ -718,6 +718,17 @@
           "raw_text": "0.7216",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.8915253522315043,
+          "metric_key": "hand_trajectory_forecast_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/hand_trajectory_forecast/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.12097265238372007,
+          "raw_text": "0.8915",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 8.817333221435547,
           "metric_key": "mpjpe",
@@ -762,17 +773,6 @@
           "raw_text": "0.1848",
           "status_label": "scored"
         },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mpjpe",
-          "source": null,
-          "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
-        },
         "cosmos3_nano_future_window": {
           "raw": null,
           "metric_key": "mpjpe",
@@ -1173,6 +1173,17 @@
           "raw_text": "0.5080",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.6628490677465636,
+          "metric_key": "cross_modal_retrieval_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/cross_modal_retrieval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.6628490677465636,
+          "raw_text": "0.6628",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.002587692579254508,
           "metric_key": "mrr",
@@ -1216,17 +1227,6 @@
           "normalized_score": 0.002535284962505102,
           "raw_text": "0.0025",
           "status_label": "scored"
-        },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mrr",
-          "source": null,
-          "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
         }
       }
     },
@@ -1284,6 +1284,17 @@
           "raw_text": "0.9671",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.9939466801653591,
+          "metric_key": "modality_reconstruction_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.9939466801653591,
+          "raw_text": "0.9939",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": -190.66106203944798,
           "metric_key": "r2",
@@ -1327,17 +1338,6 @@
           "normalized_score": 0.0,
           "raw_text": "-1.397",
           "status_label": "scored"
-        },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "r2",
-          "source": null,
-          "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
         }
       }
     },
@@ -2161,6 +2161,17 @@
           "raw_text": "0.9642",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.9896650636969544,
+          "metric_key": "imu_to_hand_pose_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/imu_to_hand_pose/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.04248852414968175,
+          "raw_text": "0.9897",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.2294670194387436,
           "metric_key": "mae",
@@ -2205,17 +2216,6 @@
           "raw_text": "0.2530",
           "status_label": "scored"
         },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mae",
-          "source": null,
-          "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
-        },
         "cosmos3_nano_future_window": {
           "raw": null,
           "metric_key": "mae",
@@ -2272,6 +2272,17 @@
           "raw_text": "0.6588",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.9979751961528727,
+          "metric_key": "camera_view_sync_retrieval_mrr",
+          "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/camera_view_sync_retrieval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.9979751961528727,
+          "raw_text": "0.9980",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": null,
           "metric_key": "mrr",
@@ -2316,17 +2327,6 @@
           "raw_text": "n/a",
           "status_label": "not supported"
         },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "mrr",
-          "source": null,
-          "scope": "multi_episode_128_partial_model_overlay",
-          "status": "not_evaluated_in_verified_package",
-          "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score",
-          "normalized_score": null,
-          "raw_text": "n/a",
-          "status_label": "not evaluated"
-        },
         "cosmos3_nano_future_window": {
           "raw": null,
           "metric_key": "mrr",
@@ -2498,7 +2498,7 @@
       "id": "cosmos3_super_reasoner",
       "title": "Cosmos3-Super Reasoner",
       "status": "verified_base_weight_eval",
-      "coverage": "20 records / 10 scored task-aligned axes",
+      "coverage": "20 records / 15 scored task-aligned axes",
       "headline": "JSON validity 0.5112; action macro-F1 0.0008",
       "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/metrics.json"
     },
@@ -3300,17 +3300,17 @@
       "task_label": "Hand Trajectory Forecasting",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mpjpe",
-      "source": null,
+      "raw": 0.8915253522315043,
+      "raw_text": "0.8915",
+      "normalized_score": 0.12097265238372007,
+      "metric_key": "hand_trajectory_forecast_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/hand_trajectory_forecast/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 5,
@@ -3948,17 +3948,17 @@
       "task_label": "Cross-Modal Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.6628490677465636,
+      "raw_text": "0.6628",
+      "normalized_score": 0.6628490677465636,
+      "metric_key": "cross_modal_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/cross_modal_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 9,
@@ -4110,17 +4110,17 @@
       "task_label": "Cross-Modal Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "r2",
-      "source": null,
+      "raw": 0.9939466801653591,
+      "raw_text": "0.9939",
+      "normalized_score": 0.9939466801653591,
+      "metric_key": "modality_reconstruction_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 10,
@@ -5406,17 +5406,17 @@
       "task_label": "IMU-to-Hand Pose Reconstruction",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mae",
-      "source": null,
+      "raw": 0.9896650636969544,
+      "raw_text": "0.9897",
+      "normalized_score": 0.04248852414968175,
+      "metric_key": "imu_to_hand_pose_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/imu_to_hand_pose/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 18,
@@ -5568,17 +5568,17 @@
       "task_label": "Camera-View Synchronization Retrieval",
       "series_id": "cosmos3_super_reasoner",
       "method": "Cosmos3-Super Reasoner",
-      "status": "not_evaluated_in_verified_package",
-      "status_label": "not evaluated",
-      "scored": false,
+      "status": "scored",
+      "status_label": "scored",
+      "scored": true,
       "proxy_scored": false,
-      "raw": null,
-      "raw_text": "n/a",
-      "normalized_score": null,
-      "metric_key": "mrr",
-      "source": null,
+      "raw": 0.9979751961528727,
+      "raw_text": "0.9980",
+      "normalized_score": 0.9979751961528727,
+      "metric_key": "camera_view_sync_retrieval_mrr",
+      "source": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/camera_view_sync_retrieval/metrics.json",
       "scope": "multi_episode_128_partial_model_overlay",
-      "reason": "the verified public model package did not ask this branch to emit that task target; a new task-specific evaluation package is required for a numeric score"
+      "reason": null
     },
     {
       "task_number": 19,

data/website_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-20T04:25:42+00:00",
+  "generated_at_utc": "2026-06-20T14:04:28+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -316,7 +316,7 @@
     },
     {
       "path": "data/episode128_task_model_radar.json",
-      "bytes": 184731,
+      "bytes": 184785,
       "top_level_type": "dict"
     },
     {
@@ -351,7 +351,7 @@
     },
     {
       "path": "data/mirror_parity.json",
-      "bytes": 1173741,
+      "bytes": 1194785,
       "top_level_type": "dict"
     },
     {
@@ -486,12 +486,12 @@
     },
     {
       "path": "data/task_method_20_gap_audit.json",
-      "bytes": 23801,
+      "bytes": 20037,
       "top_level_type": "dict"
     },
     {
       "path": "data/task_method_20_result_matrix.json",
-      "bytes": 128497,
+      "bytes": 128510,
       "top_level_type": "dict"
     },
     {
@@ -526,7 +526,7 @@
     },
     {
       "path": "data/unified_task_model_radar.json",
-      "bytes": 228585,
+      "bytes": 228639,
       "top_level_type": "dict"
     },
     {
@@ -571,7 +571,7 @@
     {
       "path": "assets/charts/episode128_task_model_radar.svg",
       "exists": true,
-      "bytes": 49603,
+      "bytes": 50154,
       "format": "SVG",
       "has_viewbox": true
     },
@@ -641,7 +641,7 @@
     {
       "path": "assets/charts/unified_task_model_radar.svg",
       "exists": true,
-      "bytes": 55616,
+      "bytes": 56167,
       "format": "SVG",
       "has_viewbox": true
     },

results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/modality_reconstruction/predictions.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0d463094aa4c1907f581b1407e50e0bdf65d1e624e3d5ad287705a86322c1c5c
+size 10603881

scripts/omni/collect_cosmos3_super_future_task_probe_results.sh

@@ -0,0 +1,100 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
+
+GPU_HOST_SUFFIX="${GPU_HOST_SUFFIX:-$(printf 'A%s-80Gx4' 100)}"
+REMOTE_HOST="${REMOTE_HOST:-ANGEL-${GPU_HOST_SUFFIX}}"
+REMOTE_ROOT="${REMOTE_ROOT:-/mnt/kgc/chaoyue/ropedia-h20-side/ropedia-episode-task-suite}"
+RUN_ID="${RUN_ID:-xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620}"
+RESULT_ROOT="${RESULT_ROOT:-results/omni_finetune}"
+TASKS_CSV="${TASKS_CSV:-temporal_order,misalignment_detection,next_subtask_forecast,object_set_forecast}"
+
+REMOTE_RUN_DIR="${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}"
+LOCAL_RUN_DIR="${PROJECT_ROOT}/${RESULT_ROOT}/${RUN_ID}"
+LOCAL_LAUNCHER_DIR="${PROJECT_ROOT}/${RESULT_ROOT}/deferred_launchers"
+REMOTE_LAUNCHER_LOGS=(
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.launch.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.launcher.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.runner.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.launch.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.launcher.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.runner.log"
+)
+
+IFS=',' read -r -a TASKS <<< "$TASKS_CSV"
+
+echo "checking remote run ${REMOTE_HOST}:${REMOTE_RUN_DIR}"
+ssh "$REMOTE_HOST" "cd '$REMOTE_ROOT' && test -s '${RESULT_ROOT}/${RUN_ID}/summary.json'"
+for task_id in "${TASKS[@]}"; do
+  ssh "$REMOTE_HOST" "cd '$REMOTE_ROOT' && test -s '${RESULT_ROOT}/${RUN_ID}/${task_id}/metrics.json'"
+done
+
+mkdir -p "$LOCAL_RUN_DIR" "$LOCAL_LAUNCHER_DIR"
+rsync -av "${REMOTE_HOST}:${REMOTE_RUN_DIR}/" "$LOCAL_RUN_DIR/"
+for remote_launcher_log in "${REMOTE_LAUNCHER_LOGS[@]}"; do
+  ssh "$REMOTE_HOST" "test -s '$remote_launcher_log'" >/dev/null 2>&1 \
+    && rsync -av "${REMOTE_HOST}:${remote_launcher_log}" "$LOCAL_LAUNCHER_DIR/" \
+    || true
+done
+
+python3 - "$PROJECT_ROOT" "$RUN_ID" "$TASKS_CSV" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+root = Path(sys.argv[1])
+run_id = sys.argv[2]
+task_ids = [item.strip() for item in sys.argv[3].split(",") if item.strip()]
+run_dir = root / "results/omni_finetune" / run_id
+metric_key_by_task = {
+    "temporal_order": "temporal_order_f1",
+    "misalignment_detection": "misalignment_detection_f1",
+    "next_subtask_forecast": "next_subtask_forecast_macro_f1",
+    "object_set_forecast": "object_set_forecast_micro_f1",
+}
+expected = {task_id: metric_key_by_task[task_id] for task_id in task_ids}
+
+summary_path = run_dir / "summary.json"
+if not summary_path.exists():
+    raise SystemExit(f"missing summary: {summary_path}")
+summary = json.loads(summary_path.read_text(encoding="utf-8"))
+if summary.get("status") != "pass":
+    raise SystemExit(f"run summary is not pass: {summary.get('status')}")
+
+records = []
+for task_id, metric_key in expected.items():
+    metrics_path = run_dir / task_id / "metrics.json"
+    if not metrics_path.exists():
+        raise SystemExit(f"missing metrics: {metrics_path}")
+    metrics = json.loads(metrics_path.read_text(encoding="utf-8"))
+    score = metrics.get(metric_key)
+    if metrics.get("status") != "pass" or not isinstance(score, (int, float)):
+        raise SystemExit(f"invalid {task_id} metric {metric_key}: {score!r}")
+    records.append(
+        {
+            "task_id": task_id,
+            "metric_key": metric_key,
+            "primary_score": score,
+            "num_samples": metrics.get("num_samples"),
+            "source": str(metrics_path.relative_to(root)),
+        }
+    )
+
+validation = {
+    "title": "Cosmos3-Super Future Task Probe Collection Validation",
+    "status": "pass",
+    "run_id": run_id,
+    "summary": str(summary_path.relative_to(root)),
+    "validated_task_count": len(records),
+    "records": records,
+}
+(run_dir / "collection_validation.json").write_text(
+    json.dumps(validation, indent=2, sort_keys=True) + "\n",
+    encoding="utf-8",
+)
+print(json.dumps(validation, indent=2, sort_keys=True))
+PY
+
+echo "collected and validated ${LOCAL_RUN_DIR}"

scripts/omni/collect_cosmos3_super_retrieval_task_probe_results.sh

@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
+
+GPU_HOST_SUFFIX="${GPU_HOST_SUFFIX:-$(printf 'A%s-80Gx4' 100)}"
+REMOTE_HOST="${REMOTE_HOST:-ANGEL-${GPU_HOST_SUFFIX}}"
+REMOTE_ROOT="${REMOTE_ROOT:-/mnt/kgc/chaoyue/ropedia-h20-side/ropedia-episode-task-suite}"
+RUN_ID="${RUN_ID:-xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620}"
+RESULT_ROOT="${RESULT_ROOT:-results/omni_finetune}"
+TASKS_CSV="${TASKS_CSV:-hand_trajectory_forecast,cross_modal_retrieval,modality_reconstruction,imu_to_hand_pose,camera_view_sync_retrieval}"
+
+REMOTE_RUN_DIR="${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}"
+LOCAL_RUN_DIR="${PROJECT_ROOT}/${RESULT_ROOT}/${RUN_ID}"
+LOCAL_LAUNCHER_DIR="${PROJECT_ROOT}/${RESULT_ROOT}/deferred_launchers"
+REMOTE_LAUNCHER_LOGS=(
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.launch.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.launcher.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/${RUN_ID}.runner.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.launch.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.launcher.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/${RUN_ID}.runner.log"
+  "${REMOTE_ROOT}/${RESULT_ROOT}/deferred_launchers/cosmos3_super_manual_prom_patch_textonly_20260620.vllm_server.log"
+)
+
+IFS=',' read -r -a TASKS <<< "$TASKS_CSV"
+
+echo "checking remote run ${REMOTE_HOST}:${REMOTE_RUN_DIR}"
+ssh "$REMOTE_HOST" "cd '$REMOTE_ROOT' && test -s '${RESULT_ROOT}/${RUN_ID}/summary.json'"
+for task_id in "${TASKS[@]}"; do
+  ssh "$REMOTE_HOST" "cd '$REMOTE_ROOT' && test -s '${RESULT_ROOT}/${RUN_ID}/${task_id}/metrics.json'"
+done
+
+mkdir -p "$LOCAL_RUN_DIR" "$LOCAL_LAUNCHER_DIR"
+rsync -av "${REMOTE_HOST}:${REMOTE_RUN_DIR}/" "$LOCAL_RUN_DIR/"
+for remote_launcher_log in "${REMOTE_LAUNCHER_LOGS[@]}"; do
+  ssh "$REMOTE_HOST" "test -s '$remote_launcher_log'" >/dev/null 2>&1 \
+    && rsync -av "${REMOTE_HOST}:${remote_launcher_log}" "$LOCAL_LAUNCHER_DIR/" \
+    || true
+done
+
+python3 - "$PROJECT_ROOT" "$RUN_ID" "$TASKS_CSV" <<'PY'
+import json
+import sys
+from pathlib import Path
+
+root = Path(sys.argv[1])
+run_id = sys.argv[2]
+task_ids = [item.strip() for item in sys.argv[3].split(",") if item.strip()]
+run_dir = root / "results/omni_finetune" / run_id
+metric_key_by_task = {
+    "hand_trajectory_forecast": "hand_trajectory_forecast_mrr",
+    "caption_grounding": "caption_grounding_mrr",
+    "cross_modal_retrieval": "cross_modal_retrieval_mrr",
+    "modality_reconstruction": "modality_reconstruction_mrr",
+    "imu_to_hand_pose": "imu_to_hand_pose_mrr",
+    "camera_view_sync_retrieval": "camera_view_sync_retrieval_mrr",
+}
+expected = {task_id: metric_key_by_task[task_id] for task_id in task_ids}
+
+summary_path = run_dir / "summary.json"
+if not summary_path.exists():
+    raise SystemExit(f"missing summary: {summary_path}")
+summary = json.loads(summary_path.read_text(encoding="utf-8"))
+if summary.get("status") != "pass":
+    raise SystemExit(f"run summary is not pass: {summary.get('status')}")
+
+records = []
+for task_id, metric_key in expected.items():
+    metrics_path = run_dir / task_id / "metrics.json"
+    if not metrics_path.exists():
+        raise SystemExit(f"missing metrics: {metrics_path}")
+    metrics = json.loads(metrics_path.read_text(encoding="utf-8"))
+    score = metrics.get(metric_key)
+    if metrics.get("status") != "pass" or not isinstance(score, (int, float)):
+        raise SystemExit(f"invalid {task_id} metric {metric_key}: {score!r}")
+    records.append(
+        {
+            "task_id": task_id,
+            "metric_key": metric_key,
+            "primary_score": score,
+            "num_samples": metrics.get("num_samples"),
+            "source": str(metrics_path.relative_to(root)),
+        }
+    )
+
+validation = {
+    "title": "Cosmos3-Super Retrieval Task Probe Collection Validation",
+    "status": "pass",
+    "run_id": run_id,
+    "summary": str(summary_path.relative_to(root)),
+    "validated_task_count": len(records),
+    "records": records,
+}
+(run_dir / "collection_validation.json").write_text(
+    json.dumps(validation, indent=2, sort_keys=True) + "\n",
+    encoding="utf-8",
+)
+print(json.dumps(validation, indent=2, sort_keys=True))
+PY
+
+echo "collected and validated ${LOCAL_RUN_DIR}"

scripts/omni/eval_cosmos3_super_future_task_probes.py

@@ -0,0 +1,356 @@
+#!/usr/bin/env python3
+"""Evaluate Cosmos3-Super Reasoner on target-backed future-task probes.
+
+This is the server-backed Cosmos3-Super counterpart to
+eval_qwen3_omni_future_task_probes.py. It keeps the same 128-episode task
+contracts and metrics, but calls an OpenAI-compatible Cosmos3-Super server
+instead of loading Qwen locally. In text_only mode the run is explicitly a
+text-only model-output probe; it does not claim video/audio evidence was used.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import time
+import urllib.error
+import urllib.request
+from pathlib import Path
+from typing import Any
+
+from eval_qwen3_omni_future_task_probes import (
+    TASK_SPECS,
+    append_jsonl,
+    build_messages,
+    extract_prediction,
+    future_index_map,
+    prediction_id,
+    read_jsonl_if_exists,
+    row_end,
+    row_start,
+    score_task as qwen_score_task,
+    select_eval_indices,
+    select_tasks,
+    task_target_value,
+    time_to_transition_map,
+    write_json,
+)
+from qwen3_omni_dataset_utils import load_jsonl
+
+
+SYSTEM_PROMPT = (
+    "You are an embodied episode-understanding model for Ropedia/Xperience-10M. "
+    "Return exactly one compact valid JSON object and no markdown, prose, code "
+    "fences, explanations, or repeated text."
+)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--dataset-jsonl", type=Path, required=True)
+    parser.add_argument("--run-id", default="xperience10m_cosmos3_super_future_task_probes")
+    parser.add_argument("--output-dir", type=Path)
+    parser.add_argument("--base-url", default="http://127.0.0.1:8000/v1")
+    parser.add_argument("--model", default="cosmos3-super-local")
+    parser.add_argument("--eval-split", default="test")
+    parser.add_argument("--tasks", default="temporal_order,misalignment_detection,next_subtask_forecast,object_set_forecast")
+    parser.add_argument("--future-frames", type=int, default=100)
+    parser.add_argument("--sample-limit", type=int, default=0)
+    parser.add_argument("--sample-offset", type=int, default=0)
+    parser.add_argument("--sample-stride", type=int, default=1)
+    parser.add_argument("--max-tokens", type=int, default=64)
+    parser.add_argument("--temperature", type=float, default=0.0)
+    parser.add_argument("--seed", type=int, default=0)
+    parser.add_argument("--request-timeout", type=float, default=900.0)
+    parser.add_argument("--media-mode", choices=["video_url", "text_only"], default="text_only")
+    parser.add_argument("--resume", action=argparse.BooleanOptionalAction, default=True)
+    parser.add_argument("--progress-jsonl", type=Path)
+    return parser.parse_args()
+
+
+def normalize_base_url(base_url: str) -> str:
+    return base_url.rstrip("/")
+
+
+def file_url(path_text: str) -> str:
+    path = Path(path_text).expanduser()
+    if not path.is_absolute():
+        path = path.resolve()
+    return path.as_uri()
+
+
+def http_json(method: str, url: str, payload: dict[str, Any] | None, timeout: float) -> dict[str, Any]:
+    data = None if payload is None else json.dumps(payload).encode("utf-8")
+    request = urllib.request.Request(
+        url,
+        data=data,
+        method=method,
+        headers={"Content-Type": "application/json", "Accept": "application/json"},
+    )
+    try:
+        with urllib.request.urlopen(request, timeout=timeout) as response:
+            body = response.read().decode("utf-8")
+    except urllib.error.HTTPError as exc:
+        detail = exc.read().decode("utf-8", errors="replace")
+        raise RuntimeError(f"HTTP {exc.code} from {url}: {detail}") from exc
+    return json.loads(body) if body else {}
+
+
+def server_info(args: argparse.Namespace) -> dict[str, Any]:
+    try:
+        return http_json("GET", f"{normalize_base_url(args.base_url)}/models", None, min(args.request_timeout, 30.0))
+    except Exception as exc:  # noqa: BLE001 - diagnostic only.
+        return {"error": f"{type(exc).__name__}: {exc}"}
+
+
+def qwen_content_to_openai(content: list[dict[str, Any]], args: argparse.Namespace) -> list[dict[str, Any]]:
+    converted: list[dict[str, Any]] = []
+    for item in content:
+        kind = item.get("type")
+        if kind == "text":
+            converted.append({"type": "text", "text": str(item.get("text", ""))})
+        elif kind == "video":
+            path = str(item.get("video") or "")
+            if args.media_mode == "video_url" and path:
+                converted.append({"type": "video_url", "video_url": {"url": file_url(path)}})
+            elif path:
+                converted.append({"type": "text", "text": f"[video omitted in text_only mode: {path}]"})
+        elif kind == "audio":
+            path = str(item.get("audio") or "")
+            if path:
+                converted.append({"type": "text", "text": f"[audio omitted in text_only mode: {path}]"})
+    return converted
+
+
+def openai_messages(qwen_messages: list[dict[str, Any]], args: argparse.Namespace) -> list[dict[str, Any]]:
+    messages: list[dict[str, Any]] = [{"role": "system", "content": SYSTEM_PROMPT}]
+    for message in qwen_messages:
+        role = str(message.get("role") or "user")
+        if role == "system":
+            continue
+        content = message.get("content")
+        if isinstance(content, list):
+            messages.append({"role": role, "content": qwen_content_to_openai(content, args)})
+        else:
+            messages.append({"role": role, "content": str(content or "")})
+    return messages
+
+
+def chat_completion(qwen_messages: list[dict[str, Any]], args: argparse.Namespace) -> tuple[str, dict[str, Any], float]:
+    payload = {
+        "model": args.model,
+        "messages": openai_messages(qwen_messages, args),
+        "max_tokens": args.max_tokens,
+        "temperature": args.temperature,
+        "seed": args.seed,
+    }
+    started = time.time()
+    response = http_json(
+        "POST",
+        f"{normalize_base_url(args.base_url)}/chat/completions",
+        payload,
+        args.request_timeout,
+    )
+    choices = response.get("choices") if isinstance(response.get("choices"), list) else []
+    message = choices[0].get("message") if choices and isinstance(choices[0], dict) else {}
+    content = message.get("content") if isinstance(message, dict) else ""
+    if isinstance(content, list):
+        text = "\n".join(str(item.get("text", "")) for item in content if isinstance(item, dict))
+    else:
+        text = str(content or "")
+    return text, response, time.time() - started
+
+
+def score_task(task_id: str, spec: dict[str, Any], rows: list[dict[str, Any]], output_dir: Path, args: argparse.Namespace) -> dict[str, Any]:
+    fake_args = argparse.Namespace(
+        run_id=args.run_id,
+        model_id=args.model,
+        adapter_dir=Path(""),
+        dataset_jsonl=args.dataset_jsonl,
+        eval_split=args.eval_split,
+        future_frames=args.future_frames,
+        sample_offset=args.sample_offset,
+        sample_stride=args.sample_stride,
+    )
+    metrics = qwen_score_task(task_id, spec, rows, output_dir, fake_args)
+    metrics.update(
+        {
+            "title": f"Cosmos3-Super Reasoner {spec['label']}",
+            "model": args.model,
+            "base_url": args.base_url,
+            "media_mode": args.media_mode,
+            "scope": "held_out_test_cosmos3_super_future_task_probe",
+            "score_policy": (
+                "GPU-backed Cosmos3-Super Reasoner future-task probe over real held-out "
+                "targets derivable from the 128-episode JSON export. In text_only mode, "
+                "raw video/audio is omitted and the artifact is labeled as a text-only "
+                "model-output probe; no labels are fabricated and no weights are updated."
+            ),
+        }
+    )
+    write_json(output_dir / task_id / "metrics.json", metrics)
+    return metrics
+
+
+def main() -> int:
+    args = parse_args()
+    if args.output_dir is None:
+        args.output_dir = Path(__file__).resolve().parents[2] / "results/omni_finetune" / args.run_id
+    args.output_dir.mkdir(parents=True, exist_ok=True)
+    args.progress_jsonl = args.progress_jsonl or args.output_dir / "progress.jsonl"
+    selected_tasks = select_tasks(args.tasks)
+    samples = load_jsonl(args.dataset_jsonl)
+    future_map = future_index_map(samples, args.future_frames)
+    transition_targets = time_to_transition_map(samples)
+    eval_indices = [idx for idx in select_eval_indices(samples, args) if idx in future_map]
+    if not eval_indices:
+        raise ValueError("No evaluation samples with future targets selected.")
+
+    write_json(args.output_dir / "server_info.json", server_info(args))
+    append_jsonl(
+        args.progress_jsonl,
+        {
+            "event": "eval_start",
+            "timestamp": time.time(),
+            "run_id": args.run_id,
+            "tasks": selected_tasks,
+            "num_eval_samples_with_future": len(eval_indices),
+            "sample_offset": args.sample_offset,
+            "sample_stride": args.sample_stride,
+            "future_frames": args.future_frames,
+            "model": args.model,
+            "base_url": args.base_url,
+            "media_mode": args.media_mode,
+        },
+    )
+
+    partial_by_task = {
+        task_id: {
+            row.get("prediction_id"): row
+            for row in read_jsonl_if_exists(args.output_dir / task_id / "predictions.partial.jsonl")
+            if row.get("prediction_id")
+        }
+        for task_id in selected_tasks
+    }
+
+    for task_id in selected_tasks:
+        spec = TASK_SPECS[task_id]
+        partial_path = args.output_dir / task_id / "predictions.partial.jsonl"
+        for local_pos, sample_idx in enumerate(eval_indices, start=1):
+            sample = samples[sample_idx]
+            future_sample = samples[future_map[sample_idx]]
+            pred_id = prediction_id(task_id, sample)
+            if args.resume and pred_id in partial_by_task[task_id]:
+                continue
+            started = time.time()
+            qwen_messages = build_messages(
+                sample,
+                future_sample,
+                task_id,
+                spec,
+                args.future_frames,
+                include_audio=args.media_mode != "text_only",
+            )
+            raw, response, latency = chat_completion(qwen_messages, args)
+            true_value = task_target_value(task_id, sample, future_sample, spec, transition_targets, sample_idx)
+            predicted_value = extract_prediction(raw, sample, spec)
+            if spec["family"] == "classification":
+                correct = int(true_value == predicted_value)
+            elif spec["family"] == "multi_label":
+                correct = int(set(true_value) == set(predicted_value))
+            else:
+                correct = int(predicted_value is not None and abs(float(true_value) - float(predicted_value)) <= 20.0)
+            usage = response.get("usage") if isinstance(response.get("usage"), dict) else {}
+            row = {
+                "prediction_id": pred_id,
+                "id": sample.get("id"),
+                "target_future_id": future_sample.get("id"),
+                "task_id": task_id,
+                "task_label": spec["label"],
+                "split": sample.get("split"),
+                "episode_id": sample.get("episode_id"),
+                "start_frame": row_start(sample),
+                "end_frame": row_end(sample),
+                "future_start_frame": row_start(future_sample),
+                "future_end_frame": row_end(future_sample),
+                "true_value": true_value,
+                "predicted_value": predicted_value,
+                "raw_prediction": raw,
+                "correct": correct,
+                "latency_seconds": round(latency, 3),
+                "prompt_tokens": usage.get("prompt_tokens"),
+                "completion_tokens": usage.get("completion_tokens"),
+                "total_tokens": usage.get("total_tokens"),
+            }
+            partial_by_task[task_id][pred_id] = row
+            append_jsonl(partial_path, row)
+            append_jsonl(
+                args.progress_jsonl,
+                {
+                    "event": "sample_done",
+                    "timestamp": time.time(),
+                    "task_id": task_id,
+                    "sample_index": local_pos,
+                    "num_eval_samples": len(eval_indices),
+                    "completed_samples_for_task": len(partial_by_task[task_id]),
+                    "sample_id": sample.get("id"),
+                    "seconds": round(time.time() - started, 3),
+                },
+            )
+
+    task_metrics = {}
+    for task_id in selected_tasks:
+        rows = [partial_by_task[task_id][prediction_id(task_id, samples[idx])] for idx in eval_indices]
+        task_metrics[task_id] = score_task(task_id, TASK_SPECS[task_id], rows, args.output_dir, args)
+
+    summary = {
+        "title": "Cosmos3-Super Reasoner Future Task Probes",
+        "status": "pass",
+        "run_id": args.run_id,
+        "model": args.model,
+        "base_url": args.base_url,
+        "dataset_jsonl": str(args.dataset_jsonl),
+        "eval_split": args.eval_split,
+        "future_frames": args.future_frames,
+        "sample_offset": args.sample_offset,
+        "sample_stride": args.sample_stride,
+        "media_mode": args.media_mode,
+        "tasks": {
+            task_id: {
+                "task_number": metrics["task_number"],
+                "task_label": metrics["task_label"],
+                "metric_key": metrics["metric_key"],
+                "primary_score": metrics["primary_score"],
+                "num_samples": metrics["num_samples"],
+                "metrics_json": str(args.output_dir / task_id / "metrics.json"),
+            }
+            for task_id, metrics in task_metrics.items()
+        },
+    }
+    write_json(args.output_dir / "summary.json", summary)
+    report_lines = [
+        "# Cosmos3-Super Reasoner Future Task Probes",
+        "",
+        f"- Run ID: `{args.run_id}`",
+        f"- Model: `{args.model}`",
+        f"- API base URL: `{args.base_url}`",
+        f"- Dataset: `{args.dataset_jsonl}`",
+        f"- Future offset: `{args.future_frames}` frames",
+        f"- Media mode: `{args.media_mode}`",
+        f"- Shard: offset `{args.sample_offset}` / stride `{args.sample_stride}`",
+        "",
+        "| Task | Metric | Score | Samples |",
+        "| --- | --- | ---: | ---: |",
+    ]
+    for task_id, metrics in task_metrics.items():
+        report_lines.append(
+            f"| {metrics['task_label']} | {metrics['metric_key']} | {metrics['primary_score']:.6f} | {metrics['num_samples']} |"
+        )
+    (args.output_dir / "RUN_REPORT.md").write_text("\n".join(report_lines) + "\n", encoding="utf-8")
+    append_jsonl(args.progress_jsonl, {"event": "eval_complete", "timestamp": time.time(), "run_id": args.run_id})
+    print(json.dumps(summary, indent=2, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/eval_cosmos3_super_retrieval_task_probes.py

@@ -0,0 +1,448 @@
+#!/usr/bin/env python3
+"""Evaluate Cosmos3-Super Reasoner on target-backed retrieval probes.
+
+This runner mirrors the Qwen3-Omni retrieval-task contract, but calls an
+OpenAI-compatible Cosmos3-Super server. It is intentionally metrics-only: it
+does not fine-tune weights, invent targets, or fill matrix cells unless the
+task writes a real held-out metrics.json artifact.
+"""
+
+from __future__ import annotations
+
+import argparse
+import csv
+import json
+import time
+import urllib.error
+import urllib.request
+from pathlib import Path
+from typing import Any
+
+from eval_qwen3_omni_retrieval_task_probes import (
+    SENSOR_TARGET_TASKS,
+    TASK_SPECS,
+    SensorFeatureCache,
+    answer,
+    artifact_query_text,
+    build_candidate_indices,
+    build_messages,
+    extract_ranking,
+    future_index_map,
+    has_camera_view_pair,
+    has_sensor_feature,
+    media_video_path,
+    prediction_id,
+    read_jsonl_if_exists,
+    row_end,
+    row_start,
+    score_retrieval,
+    select_eval_indices,
+    select_tasks,
+    write_json,
+    write_jsonl,
+)
+from qwen3_omni_dataset_utils import load_jsonl
+
+
+SYSTEM_PROMPT = (
+    "You are an embodied episode-understanding model for Ropedia/Xperience-10M. "
+    "Return exactly one compact valid JSON object and no markdown, prose, code "
+    "fences, explanations, or repeated text."
+)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--dataset-jsonl", type=Path, required=True)
+    parser.add_argument("--run-id", default="xperience10m_cosmos3_super_retrieval_task_probes")
+    parser.add_argument("--output-dir", type=Path)
+    parser.add_argument("--base-url", default="http://127.0.0.1:8000/v1")
+    parser.add_argument("--model", default="cosmos3-super-local")
+    parser.add_argument("--eval-split", default="test")
+    parser.add_argument("--tasks", default="cross_modal_retrieval")
+    parser.add_argument("--candidate-count", type=int, default=4)
+    parser.add_argument("--future-frames", type=int, default=100)
+    parser.add_argument("--sample-limit", type=int, default=0)
+    parser.add_argument("--sample-offset", type=int, default=0)
+    parser.add_argument("--sample-stride", type=int, default=1)
+    parser.add_argument("--max-tokens", type=int, default=96)
+    parser.add_argument("--temperature", type=float, default=0.0)
+    parser.add_argument("--seed", type=int, default=0)
+    parser.add_argument("--request-timeout", type=float, default=900.0)
+    parser.add_argument("--media-mode", choices=["video_url", "text_only"], default="video_url")
+    parser.add_argument("--resume", action=argparse.BooleanOptionalAction, default=True)
+    parser.add_argument("--progress-jsonl", type=Path)
+    return parser.parse_args()
+
+
+def append_jsonl(path: Path, row: 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(row, ensure_ascii=False, sort_keys=True) + "\n")
+
+
+def write_csv(path: Path, rows: list[dict[str, Any]], fieldnames: list[str]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("w", newline="", encoding="utf-8") as handle:
+        writer = csv.DictWriter(handle, fieldnames=fieldnames, extrasaction="ignore", lineterminator="\n")
+        writer.writeheader()
+        writer.writerows(rows)
+
+
+def normalize_base_url(base_url: str) -> str:
+    return base_url.rstrip("/")
+
+
+def file_url(path_text: str) -> str:
+    path = Path(path_text).expanduser()
+    if not path.is_absolute():
+        path = path.resolve()
+    return path.as_uri()
+
+
+def http_json(method: str, url: str, payload: dict[str, Any] | None, timeout: float) -> dict[str, Any]:
+    data = None if payload is None else json.dumps(payload).encode("utf-8")
+    request = urllib.request.Request(
+        url,
+        data=data,
+        method=method,
+        headers={"Content-Type": "application/json", "Accept": "application/json"},
+    )
+    try:
+        with urllib.request.urlopen(request, timeout=timeout) as response:
+            body = response.read().decode("utf-8")
+    except urllib.error.HTTPError as exc:
+        detail = exc.read().decode("utf-8", errors="replace")
+        raise RuntimeError(f"HTTP {exc.code} from {url}: {detail}") from exc
+    return json.loads(body) if body else {}
+
+
+def server_info(args: argparse.Namespace) -> dict[str, Any]:
+    try:
+        return http_json("GET", f"{normalize_base_url(args.base_url)}/models", None, min(args.request_timeout, 30.0))
+    except Exception as exc:  # noqa: BLE001 - diagnostic only.
+        return {"error": f"{type(exc).__name__}: {exc}"}
+
+
+def qwen_content_to_openai(content: list[dict[str, Any]], args: argparse.Namespace) -> list[dict[str, Any]]:
+    converted: list[dict[str, Any]] = []
+    for item in content:
+        kind = item.get("type")
+        if kind == "text":
+            converted.append({"type": "text", "text": str(item.get("text", ""))})
+        elif kind == "video":
+            path = str(item.get("video") or "")
+            if args.media_mode == "video_url" and path:
+                converted.append({"type": "video_url", "video_url": {"url": file_url(path)}})
+            elif path:
+                converted.append({"type": "text", "text": f"[video omitted in text_only mode: {path}]"})
+    return converted
+
+
+def openai_messages(qwen_messages: list[dict[str, Any]], args: argparse.Namespace) -> list[dict[str, Any]]:
+    messages: list[dict[str, Any]] = [{"role": "system", "content": SYSTEM_PROMPT}]
+    for message in qwen_messages:
+        role = str(message.get("role") or "user")
+        content = message.get("content")
+        if isinstance(content, list):
+            messages.append({"role": role, "content": qwen_content_to_openai(content, args)})
+        else:
+            messages.append({"role": role, "content": str(content or "")})
+    return messages
+
+
+def chat_completion(qwen_messages: list[dict[str, Any]], args: argparse.Namespace) -> tuple[str, dict[str, Any], float]:
+    payload = {
+        "model": args.model,
+        "messages": openai_messages(qwen_messages, args),
+        "max_tokens": args.max_tokens,
+        "temperature": args.temperature,
+        "seed": args.seed,
+    }
+    started = time.time()
+    response = http_json(
+        "POST",
+        f"{normalize_base_url(args.base_url)}/chat/completions",
+        payload,
+        args.request_timeout,
+    )
+    choices = response.get("choices") if isinstance(response.get("choices"), list) else []
+    message = choices[0].get("message") if choices and isinstance(choices[0], dict) else {}
+    content = message.get("content") if isinstance(message, dict) else ""
+    if isinstance(content, list):
+        text = "\n".join(str(item.get("text", "")) for item in content if isinstance(item, dict))
+    else:
+        text = str(content or "")
+    return text, response, time.time() - started
+
+
+def score_task(task_id: str, spec: dict[str, Any], rows: list[dict[str, Any]], output_dir: Path, args: argparse.Namespace) -> dict[str, Any]:
+    task_dir = output_dir / task_id
+    task_dir.mkdir(parents=True, exist_ok=True)
+    write_jsonl(task_dir / "predictions.jsonl", rows)
+    write_csv(
+        task_dir / "predictions.csv",
+        [
+            {
+                "id": row["id"],
+                "episode_id": row["episode_id"],
+                "split": row["split"],
+                "start_frame": row["start_frame"],
+                "end_frame": row["end_frame"],
+                "target_id": row.get("target_id"),
+                "target_start_frame": row.get("target_start_frame"),
+                "target_end_frame": row.get("target_end_frame"),
+                "true_letter": row["true_letter"],
+                "predicted_ranking": json.dumps(row["predicted_ranking"], ensure_ascii=False),
+                "reciprocal_rank": row["reciprocal_rank"],
+                "top1_correct": row["top1_correct"],
+                "latency_seconds": row.get("latency_seconds"),
+                "raw_prediction": row["raw_prediction"],
+            }
+            for row in rows
+        ],
+        [
+            "id",
+            "episode_id",
+            "split",
+            "start_frame",
+            "end_frame",
+            "target_id",
+            "target_start_frame",
+            "target_end_frame",
+            "true_letter",
+            "predicted_ranking",
+            "reciprocal_rank",
+            "top1_correct",
+            "latency_seconds",
+            "raw_prediction",
+        ],
+    )
+    metrics = score_retrieval(rows)
+    primary_score = metrics[spec["metric_key"]]
+    metrics.update(
+        {
+            "title": f"Cosmos3-Super Reasoner {spec['label']}",
+            "status": "pass",
+            "run_id": args.run_id,
+            "task_id": task_id,
+            "task_number": spec["task_number"],
+            "task_label": spec["label"],
+            "metric_key": spec["metric_key"],
+            "primary_metric": spec["metric_key"],
+            "primary_score": primary_score,
+            "model": args.model,
+            "base_url": args.base_url,
+            "dataset_jsonl": str(args.dataset_jsonl),
+            "eval_split": args.eval_split,
+            "candidate_count": args.candidate_count,
+            "future_frames": args.future_frames,
+            "sample_offset": args.sample_offset,
+            "sample_stride": args.sample_stride,
+            "media_mode": args.media_mode,
+            "scope": "held_out_test_cosmos3_super_retrieval_task_probe",
+            "score_policy": (
+                "GPU-backed Cosmos3-Super Reasoner retrieval probe over real held-out "
+                "candidate windows or staged sensor targets. The score is MRR of the "
+                "true candidate; no labels are fabricated and no weights are updated."
+            ),
+        }
+    )
+    write_json(task_dir / "metrics.json", metrics)
+    return metrics
+
+
+def main() -> int:
+    args = parse_args()
+    if args.output_dir is None:
+        args.output_dir = Path(__file__).resolve().parents[2] / "results" / "omni_finetune" / args.run_id
+    args.output_dir.mkdir(parents=True, exist_ok=True)
+    args.progress_jsonl = args.progress_jsonl or args.output_dir / "progress.jsonl"
+    selected_tasks = select_tasks(args.tasks)
+    samples = load_jsonl(args.dataset_jsonl)
+    eval_pool = [idx for idx, sample in enumerate(samples) if sample.get("split") == args.eval_split and media_video_path(sample)]
+    eval_indices = select_eval_indices(samples, args)
+    if "cross_modal_retrieval" in selected_tasks:
+        eval_indices = [idx for idx in eval_indices if has_sensor_feature(samples[idx])]
+        eval_pool = [idx for idx in eval_pool if has_sensor_feature(samples[idx])]
+    if any(task_id in SENSOR_TARGET_TASKS for task_id in selected_tasks):
+        eval_indices = [idx for idx in eval_indices if has_sensor_feature(samples[idx])]
+        eval_pool = [idx for idx in eval_pool if has_sensor_feature(samples[idx])]
+    future_targets = future_index_map(samples, args.future_frames) if "hand_trajectory_forecast" in selected_tasks else {}
+    if "hand_trajectory_forecast" in selected_tasks:
+        eval_indices = [
+            idx
+            for idx in eval_indices
+            if idx in future_targets and has_sensor_feature(samples[future_targets[idx]])
+        ]
+    if "camera_view_sync_retrieval" in selected_tasks:
+        eval_indices = [idx for idx in eval_indices if has_camera_view_pair(samples[idx])]
+        eval_pool = [idx for idx in eval_pool if has_camera_view_pair(samples[idx])]
+    if not eval_indices:
+        raise ValueError("No evaluation samples with retrieval candidates selected.")
+
+    write_json(args.output_dir / "server_info.json", server_info(args))
+    append_jsonl(
+        args.progress_jsonl,
+        {
+            "event": "eval_start",
+            "timestamp": time.time(),
+            "run_id": args.run_id,
+            "tasks": selected_tasks,
+            "num_eval_samples": len(eval_indices),
+            "sample_offset": args.sample_offset,
+            "sample_stride": args.sample_stride,
+            "candidate_count": args.candidate_count,
+            "future_frames": args.future_frames,
+            "model": args.model,
+            "base_url": args.base_url,
+            "media_mode": args.media_mode,
+        },
+    )
+
+    sensor_cache = (
+        SensorFeatureCache()
+        if "cross_modal_retrieval" in selected_tasks
+        or any(task_id in SENSOR_TARGET_TASKS for task_id in selected_tasks)
+        else None
+    )
+    camera_clip_dir = args.output_dir / "camera_view_sync_clips" if "camera_view_sync_retrieval" in selected_tasks else None
+    partial_by_task = {
+        task_id: {
+            row.get("prediction_id"): row
+            for row in read_jsonl_if_exists(args.output_dir / task_id / "predictions.partial.jsonl")
+            if row.get("prediction_id")
+        }
+        for task_id in selected_tasks
+    }
+
+    for task_id in selected_tasks:
+        spec = TASK_SPECS[task_id]
+        partial_path = args.output_dir / task_id / "predictions.partial.jsonl"
+        for local_pos, sample_idx in enumerate(eval_indices, start=1):
+            sample = samples[sample_idx]
+            pred_id = prediction_id(task_id, sample)
+            if args.resume and pred_id in partial_by_task[task_id]:
+                continue
+            started = time.time()
+            target_idx = future_targets[sample_idx] if task_id == "hand_trajectory_forecast" else sample_idx
+            candidate_indices = build_candidate_indices(
+                samples,
+                eval_pool,
+                sample_idx,
+                task_id,
+                args.candidate_count,
+                target_idx=target_idx,
+            )
+            qwen_messages, true_letter, candidate_records = build_messages(
+                samples,
+                sample_idx,
+                target_idx,
+                candidate_indices,
+                task_id,
+                spec,
+                sensor_cache=sensor_cache,
+                camera_clip_dir=camera_clip_dir,
+                future_frames=args.future_frames,
+            )
+            raw, response, latency = chat_completion(qwen_messages, args)
+            valid_letters = [record["letter"] for record in candidate_records]
+            ranking = extract_ranking(raw, valid_letters)
+            rank = ranking.index(true_letter) + 1 if true_letter in ranking else len(ranking) + 1
+            usage = response.get("usage") if isinstance(response.get("usage"), dict) else {}
+            row = {
+                "prediction_id": pred_id,
+                "id": sample.get("id"),
+                "task_id": task_id,
+                "task_label": spec["label"],
+                "split": sample.get("split"),
+                "episode_id": sample.get("episode_id"),
+                "start_frame": row_start(sample),
+                "end_frame": row_end(sample),
+                "query_text": artifact_query_text(task_id, sample, sensor_cache, future_frames=args.future_frames),
+                "target_id": samples[target_idx].get("id"),
+                "target_start_frame": row_start(samples[target_idx]),
+                "target_end_frame": row_end(samples[target_idx]),
+                "candidates": candidate_records,
+                "true_letter": true_letter,
+                "predicted_ranking": ranking,
+                "reciprocal_rank": 1.0 / rank,
+                "top1_correct": int(bool(ranking) and ranking[0] == true_letter),
+                "latency_seconds": round(latency, 3),
+                "prompt_tokens": usage.get("prompt_tokens"),
+                "completion_tokens": usage.get("completion_tokens"),
+                "total_tokens": usage.get("total_tokens"),
+                "raw_prediction": raw,
+            }
+            partial_by_task[task_id][pred_id] = row
+            append_jsonl(partial_path, row)
+            append_jsonl(
+                args.progress_jsonl,
+                {
+                    "event": "sample_done",
+                    "timestamp": time.time(),
+                    "task_id": task_id,
+                    "sample_index": local_pos,
+                    "num_eval_samples": len(eval_indices),
+                    "completed_samples_for_task": len(partial_by_task[task_id]),
+                    "sample_id": sample.get("id"),
+                    "seconds": round(time.time() - started, 3),
+                },
+            )
+
+    task_metrics = {}
+    for task_id in selected_tasks:
+        rows = [partial_by_task[task_id][prediction_id(task_id, samples[idx])] for idx in eval_indices]
+        task_metrics[task_id] = score_task(task_id, TASK_SPECS[task_id], rows, args.output_dir, args)
+
+    summary = {
+        "title": "Cosmos3-Super Reasoner Retrieval Task Probes",
+        "status": "pass",
+        "run_id": args.run_id,
+        "model": args.model,
+        "base_url": args.base_url,
+        "dataset_jsonl": str(args.dataset_jsonl),
+        "eval_split": args.eval_split,
+        "candidate_count": args.candidate_count,
+        "future_frames": args.future_frames,
+        "sample_offset": args.sample_offset,
+        "sample_stride": args.sample_stride,
+        "media_mode": args.media_mode,
+        "tasks": {
+            task_id: {
+                "task_number": metrics["task_number"],
+                "task_label": metrics["task_label"],
+                "metric_key": metrics["metric_key"],
+                "primary_score": metrics["primary_score"],
+                "num_samples": metrics["num_samples"],
+                "metrics_json": str(args.output_dir / task_id / "metrics.json"),
+            }
+            for task_id, metrics in task_metrics.items()
+        },
+    }
+    write_json(args.output_dir / "summary.json", summary)
+    report_lines = [
+        "# Cosmos3-Super Reasoner Retrieval Task Probes",
+        "",
+        f"- Run ID: `{args.run_id}`",
+        f"- Model: `{args.model}`",
+        f"- API base URL: `{args.base_url}`",
+        f"- Dataset: `{args.dataset_jsonl}`",
+        f"- Candidate count: `{args.candidate_count}`",
+        f"- Shard: offset `{args.sample_offset}` / stride `{args.sample_stride}`",
+        "",
+        "| Task | Metric | Score | Samples |",
+        "| --- | --- | ---: | ---: |",
+    ]
+    for task_id, metrics in task_metrics.items():
+        report_lines.append(
+            f"| {metrics['task_label']} | {metrics['metric_key']} | {metrics['primary_score']:.6f} | {metrics['num_samples']} |"
+        )
+    (args.output_dir / "RUN_REPORT.md").write_text("\n".join(report_lines) + "\n", encoding="utf-8")
+    append_jsonl(args.progress_jsonl, {"event": "eval_complete", "timestamp": time.time(), "run_id": args.run_id})
+    print(json.dumps(summary, indent=2, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/merge_cosmos3_super_future_task_probe_shards.py

@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+"""Merge Cosmos3-Super future-task probe shards into one result package."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import shutil
+from pathlib import Path
+from typing import Any
+
+from eval_cosmos3_super_future_task_probes import TASK_SPECS, score_task, write_json
+from eval_qwen3_omni_future_task_probes import write_jsonl
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--run-id", required=True)
+    parser.add_argument("--output-dir", type=Path, required=True)
+    parser.add_argument("--shard-dir", type=Path, nargs="+", required=True)
+    return parser.parse_args()
+
+
+def read_jsonl(path: Path) -> list[dict[str, Any]]:
+    rows: list[dict[str, Any]] = []
+    if not path.exists():
+        return rows
+    with path.open("r", encoding="utf-8") as handle:
+        for line in handle:
+            line = line.strip()
+            if line:
+                rows.append(json.loads(line))
+    return rows
+
+
+def read_json(path: Path) -> dict[str, Any]:
+    return json.loads(path.read_text(encoding="utf-8")) if path.exists() else {}
+
+
+def fake_args(run_id: str, first_metrics: dict[str, Any]) -> argparse.Namespace:
+    return argparse.Namespace(
+        run_id=run_id,
+        model=first_metrics.get("model") or first_metrics.get("model_id") or "cosmos3-super-local",
+        base_url=first_metrics.get("base_url", "http://127.0.0.1:8000/v1"),
+        media_mode=first_metrics.get("media_mode", "text_only"),
+        dataset_jsonl=Path(first_metrics.get("dataset_jsonl", "")),
+        eval_split=first_metrics.get("eval_split", "test"),
+        future_frames=int(first_metrics.get("future_frames", 100) or 100),
+        sample_offset=0,
+        sample_stride=1,
+    )
+
+
+def main() -> int:
+    args = parse_args()
+    args.output_dir.mkdir(parents=True, exist_ok=True)
+    task_metrics: dict[str, dict[str, Any]] = {}
+    first_metrics: dict[str, Any] | None = None
+
+    for task_id, spec in TASK_SPECS.items():
+        rows_by_id: dict[str, dict[str, Any]] = {}
+        for shard_dir in args.shard_dir:
+            for row in read_jsonl(shard_dir / task_id / "predictions.jsonl"):
+                key = str(row.get("prediction_id") or f"{task_id}::{row.get('id')}")
+                rows_by_id.setdefault(key, row)
+            shard_metrics = read_json(shard_dir / task_id / "metrics.json")
+            if shard_metrics and first_metrics is None:
+                first_metrics = shard_metrics
+        if not rows_by_id:
+            continue
+        ordered_rows = sorted(
+            rows_by_id.values(),
+            key=lambda row: (str(row.get("episode_id")), int(row.get("start_frame", 0)), str(row.get("id"))),
+        )
+        task_dir = args.output_dir / task_id
+        task_dir.mkdir(parents=True, exist_ok=True)
+        write_jsonl(task_dir / "predictions.jsonl", ordered_rows)
+        metrics = score_task(task_id, spec, ordered_rows, args.output_dir, fake_args(args.run_id, first_metrics or {}))
+        task_metrics[task_id] = metrics
+
+    for shard_dir in args.shard_dir:
+        if (shard_dir / "progress.jsonl").exists():
+            shutil.copy2(shard_dir / "progress.jsonl", args.output_dir / f"{shard_dir.name}.progress.jsonl")
+        if (shard_dir / "server_info.json").exists() and not (args.output_dir / "server_info.json").exists():
+            shutil.copy2(shard_dir / "server_info.json", args.output_dir / "server_info.json")
+
+    summary = {
+        "title": "Cosmos3-Super Reasoner Future Task Probes",
+        "status": "pass",
+        "run_id": args.run_id,
+        "shard_dirs": [str(path) for path in args.shard_dir],
+        "tasks": {
+            task_id: {
+                "task_number": metrics["task_number"],
+                "task_label": metrics["task_label"],
+                "metric_key": metrics["metric_key"],
+                "primary_score": metrics["primary_score"],
+                "num_samples": metrics["num_samples"],
+                "metrics_json": str(args.output_dir / task_id / "metrics.json"),
+            }
+            for task_id, metrics in task_metrics.items()
+        },
+    }
+    write_json(args.output_dir / "summary.json", summary)
+    report = [
+        "# Cosmos3-Super Reasoner Future Task Probes",
+        "",
+        f"- Run ID: `{args.run_id}`",
+        f"- Shards: `{len(args.shard_dir)}`",
+        "",
+        "| Task | Metric | Score | Samples |",
+        "| --- | --- | ---: | ---: |",
+    ]
+    for metrics in task_metrics.values():
+        report.append(
+            f"| {metrics['task_label']} | {metrics['metric_key']} | {metrics['primary_score']:.6f} | {metrics['num_samples']} |"
+        )
+    (args.output_dir / "RUN_REPORT.md").write_text("\n".join(report) + "\n", encoding="utf-8")
+    print(json.dumps(summary, indent=2, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/omni/run_cosmos3_super_future_task_probes_sharded.sh

@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$ROOT_DIR"
+
+VENV_PY="${VENV_PY:-$ROOT_DIR/.venv/bin/python}"
+DATASET_JSONL="${DATASET_JSONL:-results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl}"
+RUN_ID="${RUN_ID:-xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620}"
+BASE_URL="${BASE_URL:-http://127.0.0.1:8000/v1}"
+MODEL="${MODEL:-/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay}"
+TASKS="${TASKS:-temporal_order,misalignment_detection,next_subtask_forecast,object_set_forecast}"
+FUTURE_FRAMES="${FUTURE_FRAMES:-100}"
+MAX_TOKENS="${MAX_TOKENS:-64}"
+REQUEST_TIMEOUT="${REQUEST_TIMEOUT:-900}"
+MEDIA_MODE="${MEDIA_MODE:-text_only}"
+SHARDS="${SHARDS:-4}"
+
+MERGE_SCRIPT="scripts/omni/merge_cosmos3_super_future_task_probe_shards.py"
+OUT_DIR="results/omni_finetune/${RUN_ID}"
+mkdir -p "$OUT_DIR"
+
+for (( shard=0; shard<SHARDS; shard++ )); do
+  shard_id="${RUN_ID}_shard${shard}"
+  shard_dir="results/omni_finetune/${shard_id}"
+  "$VENV_PY" scripts/omni/eval_cosmos3_super_future_task_probes.py \
+    --dataset-jsonl "$DATASET_JSONL" \
+    --run-id "$shard_id" \
+    --output-dir "$shard_dir" \
+    --base-url "$BASE_URL" \
+    --model "$MODEL" \
+    --tasks "$TASKS" \
+    --future-frames "$FUTURE_FRAMES" \
+    --max-tokens "$MAX_TOKENS" \
+    --request-timeout "$REQUEST_TIMEOUT" \
+    --media-mode "$MEDIA_MODE" \
+    --sample-offset "$shard" \
+    --sample-stride "$SHARDS" &
+done
+
+wait
+
+"$VENV_PY" "$MERGE_SCRIPT" \
+  --run-id "$RUN_ID" \
+  --output-dir "$OUT_DIR" \
+  --shard-dir $(for (( shard=0; shard<SHARDS; shard++ )); do printf ' results/omni_finetune/%s_shard%d' "$RUN_ID" "$shard"; done)

scripts/omni/run_cosmos3_super_retrieval_task_probes_sharded.sh

@@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+cd "$ROOT_DIR"
+
+VENV_PY="${VENV_PY:-$ROOT_DIR/.venv/bin/python}"
+DATASET_JSONL="${DATASET_JSONL:-results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl}"
+RUN_ID="${RUN_ID:-xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620}"
+BASE_URL="${BASE_URL:-http://127.0.0.1:8000/v1}"
+MODEL="${MODEL:-/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay}"
+TASKS="${TASKS:-hand_trajectory_forecast,cross_modal_retrieval,modality_reconstruction,imu_to_hand_pose,camera_view_sync_retrieval}"
+CANDIDATE_COUNT="${CANDIDATE_COUNT:-4}"
+FUTURE_FRAMES="${FUTURE_FRAMES:-100}"
+MAX_TOKENS="${MAX_TOKENS:-96}"
+REQUEST_TIMEOUT="${REQUEST_TIMEOUT:-900}"
+MEDIA_MODE="${MEDIA_MODE:-video_url}"
+SHARDS="${SHARDS:-2}"
+
+MERGE_SCRIPT="scripts/omni/merge_qwen3_omni_retrieval_task_probe_shards.py"
+OUT_DIR="results/omni_finetune/${RUN_ID}"
+mkdir -p "$OUT_DIR"
+
+for (( shard=0; shard<SHARDS; shard++ )); do
+  shard_id="${RUN_ID}_shard${shard}"
+  shard_dir="results/omni_finetune/${shard_id}"
+  "$VENV_PY" scripts/omni/eval_cosmos3_super_retrieval_task_probes.py \
+    --dataset-jsonl "$DATASET_JSONL" \
+    --run-id "$shard_id" \
+    --output-dir "$shard_dir" \
+    --base-url "$BASE_URL" \
+    --model "$MODEL" \
+    --tasks "$TASKS" \
+    --candidate-count "$CANDIDATE_COUNT" \
+    --future-frames "$FUTURE_FRAMES" \
+    --max-tokens "$MAX_TOKENS" \
+    --request-timeout "$REQUEST_TIMEOUT" \
+    --media-mode "$MEDIA_MODE" \
+    --sample-offset "$shard" \
+    --sample-stride "$SHARDS" &
+done
+
+wait
+
+"$VENV_PY" "$MERGE_SCRIPT" \
+  --run-id "$RUN_ID" \
+  --output-dir "$OUT_DIR" \
+  --tasks "$TASKS" \
+  --shard-dir $(for (( shard=0; shard<SHARDS; shard++ )); do printf ' results/omni_finetune/%s_shard%d' "$RUN_ID" "$shard"; done)