Add files using upload-large-folder tool

data/artifact_index.json

@@ -1,6 +1,6 @@
 {
   "title": "Ropedia Xperience-10M Task Suite Artifact Index",
-  "generated_at_utc": "2026-06-20T14:03:31+00:00",
+  "generated_at_utc": "2026-06-20T15:20:34+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": "03e110008eda658d725dc690ababf5e07dab8bab7ec4a6155d237ab9f9f95ecc"
+      "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
     },
     {
       "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": 228639,
-      "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+      "bytes": 228423,
+      "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
     },
     {
       "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": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+      "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
     },
     {
       "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": 184785,
-      "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+      "bytes": 184569,
+      "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
     },
     {
       "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": 128510,
-      "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+      "bytes": 128400,
+      "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
     },
     {
       "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": 3771,
-      "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
+      "bytes": 3739,
+      "sha256": "325df952cc73a1a530781c3f55cf87105a178ac3cb3553515c817d00916deb27"
     },
     {
       "id": "task_method_20_gap_audit_json",
@@ -786,7 +786,7 @@
       "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": 20037,
-      "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+      "sha256": "20752c326190b793e2e89415a01c29c140a0ab1bff45825cdfbcca81fe8fcdec"
     },
     {
       "id": "task_method_20_gap_audit",
@@ -797,7 +797,7 @@
       "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": 6398,
-      "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
+      "sha256": "ab8ccca20391194c3211ec3013fb0ec7f88bba1d6b94aadd46a5de3f0f5f3324"
     },
     {
       "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": 56167,
-      "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+      "bytes": 56612,
+      "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
     },
     {
       "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": 50154,
-      "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+      "bytes": 50599,
+      "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
     },
     {
       "id": "unified_task_model_radar_builder",
@@ -1116,7 +1116,7 @@
       "shows": "Machine-readable release-check summary for validators, mirrors, and public project surfaces.",
       "exists": true,
       "bytes": 8100,
-      "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+      "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
     },
     {
       "id": "public_surface_qa",
@@ -1241,7 +1241,7 @@
       "volatile": true,
       "shows": "Records the last live GitHub/HF URL verification after upload.",
       "exists": true,
-      "bytes": 172703,
+      "bytes": 181991,
       "hash_policy": "existence_and_size_only"
     },
     {
@@ -1297,7 +1297,7 @@
       "volatile": true,
       "shows": "Confirms public bundles exclude raw data, caches, heavy archives, and credential text.",
       "exists": true,
-      "bytes": 9758,
+      "bytes": 10277,
       "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": 1194785,
+      "bytes": 1231414,
       "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": 20022,
+      "bytes": 20023,
       "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": "2026-06-20T13:58:04+00:00",
+  "generated_at_utc": "2026-06-20T15:20:32+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": 124,
+  "scored_method_task_count": 128,
   "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",
@@ -149,17 +149,17 @@
       "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": 15,
-      "covered_task_count": 15,
+      "scored_task_count": 19,
+      "covered_task_count": 19,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 5,
+      "scoreless_task_count": 1,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 5,
+      "not_evaluated_task_count": 1,
       "status_counts": {
-        "not_evaluated_in_verified_package": 5,
-        "scored": 15
+        "not_evaluated_in_verified_package": 1,
+        "scored": 19
       },
-      "coverage_fraction": 0.75,
+      "coverage_fraction": 0.95,
       "result_record_fraction": 1.0
     },
     {
@@ -1166,15 +1166,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "f1",
-          "source": null,
+          "raw": 0.6286317274823326,
+          "metric_key": "temporal_order_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 0.6286317274823326,
+          "raw_text": "0.6286",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": null,
@@ -1257,15 +1257,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "f1",
-          "source": null,
+          "raw": 0.37271645981034185,
+          "metric_key": "misalignment_detection_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 0.37271645981034185,
+          "raw_text": "0.3727",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": null,
@@ -1439,15 +1439,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "macro_f1",
-          "source": null,
+          "raw": 0.0,
+          "metric_key": "next_subtask_forecast_macro_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 0.0,
+          "raw_text": "0.0000",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": 0.006614876224708678,
@@ -1712,15 +1712,15 @@
           "status_label": "scored"
         },
         "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "micro_f1",
-          "source": null,
+          "raw": 0.0009279881217520415,
+          "metric_key": "object_set_forecast_micro_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 0.0009279881217520415,
+          "raw_text": "0.0009",
+          "status_label": "scored"
         },
         "cosmos3_nano_future_window": {
           "raw": 0.01781970649895178,
@@ -3366,17 +3366,17 @@
       "task_label": "Temporal Order Verification",
       "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": "f1",
-      "source": null,
+      "raw": 0.6286317274823326,
+      "raw_text": "0.6286",
+      "normalized_score": 0.6286317274823326,
+      "metric_key": "temporal_order_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 11,
@@ -3492,17 +3492,17 @@
       "task_label": "Multimodal Synchronization Detection",
       "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": "f1",
-      "source": null,
+      "raw": 0.37271645981034185,
+      "raw_text": "0.3727",
+      "normalized_score": 0.37271645981034185,
+      "metric_key": "misalignment_detection_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 12,
@@ -3744,17 +3744,17 @@
       "task_label": "Long-Horizon Next-Subtask 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": "macro_f1",
-      "source": null,
+      "raw": 0.0,
+      "raw_text": "0.0000",
+      "normalized_score": 0.0,
+      "metric_key": "next_subtask_forecast_macro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 14,
@@ -4122,17 +4122,17 @@
       "task_label": "Future Object-Set 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": "micro_f1",
-      "source": null,
+      "raw": 0.0009279881217520415,
+      "raw_text": "0.0009",
+      "normalized_score": 0.0009279881217520415,
+      "metric_key": "object_set_forecast_micro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 17,

data/mirror_parity.json

@@ -1,9 +1,9 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:52:37+00:00",
+  "generated_at_utc": "2026-06-20T15:23:41+00:00",
   "hf_root": "hf_publish",
   "summary": {
-    "group_count": 1088,
+    "group_count": 1114,
     "failure_count": 0,
     "failures_by_surface": {}
   },
@@ -138,45 +138,45 @@
       "local": {
         "path": "repo:docs/data/artifact_index.json",
         "exists": true,
-        "bytes": 116644,
-        "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+        "bytes": 116645,
+        "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         },
         "hf_model_data": {
           "path": "hf_model:data/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         },
         "hf_model": {
           "path": "hf_model:metrics/artifact_index.json",
           "exists": true,
-          "bytes": 116644,
-          "sha256": "b5dc98a747e85ce2b82fb81f72d5ae53d71817b365c26d2cc08a2bbcdc6a098e"
+          "bytes": 116645,
+          "sha256": "822a54d5f0c4b90c582057bc8c83f22b4f9d78aaf64262c6302cf40898f42b78"
         }
       },
       "failures": []
@@ -432,45 +432,45 @@
       "local": {
         "path": "repo:docs/data/live_publication_status.json",
         "exists": true,
-        "bytes": 181998,
-        "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+        "bytes": 181991,
+        "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         },
         "hf_model_data": {
           "path": "hf_model:data/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         },
         "hf_model": {
           "path": "hf_model:metrics/live_publication_status.json",
           "exists": true,
-          "bytes": 181998,
-          "sha256": "bc2156e09848b718fbfc6a940f9bcab88fc9b6e9d1887504518acf16d367db3d"
+          "bytes": 181991,
+          "sha256": "6b6a6c3a7dca04546fbf26e9dfdbb29d5318084d7d1553018d345d83c0414cc3"
         }
       },
       "failures": []
@@ -923,44 +923,44 @@
         "path": "repo:docs/data/publication_audit.json",
         "exists": true,
         "bytes": 10277,
-        "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+        "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         },
         "hf_model_data": {
           "path": "hf_model:data/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         },
         "hf_model": {
           "path": "hf_model:metrics/publication_audit.json",
           "exists": true,
           "bytes": 10277,
-          "sha256": "f1f9b13d031dfb095c257e0457ef0871c09057baefd080c5843569111ebf50b3"
+          "sha256": "622756fbfd302077bea53ba1aaf2f210ea29ea14b3e161de3d9653f519213516"
         }
       },
       "failures": []
@@ -972,44 +972,44 @@
         "path": "repo:docs/data/public_surface_qa.json",
         "exists": true,
         "bytes": 7126,
-        "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+        "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         },
         "hf_model_data": {
           "path": "hf_model:data/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         },
         "hf_model": {
           "path": "hf_model:metrics/public_surface_qa.json",
           "exists": true,
           "bytes": 7126,
-          "sha256": "ee3b7875c232434650beef885c3c5537b2657039b71004d555e87fd8bff166df"
+          "sha256": "057cd4aef556cdf667f4c92ba67009ea0dd8f08eae9f81019a51773c5b63dea4"
         }
       },
       "failures": []
@@ -1119,44 +1119,44 @@
         "path": "repo:docs/data/quality_gates.json",
         "exists": true,
         "bytes": 8100,
-        "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+        "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         },
         "hf_model_data": {
           "path": "hf_model:data/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         },
         "hf_model": {
           "path": "hf_model:metrics/quality_gates.json",
           "exists": true,
           "bytes": 8100,
-          "sha256": "ee5c7a273f492e6fc2fe2ab92f7eae6a1a8bedf963b5faecdd362ac5495864cf"
+          "sha256": "54867eb8abcebb9948f3a6f2f5f3c3a995373b856f00d1af23212a7c39633bc4"
         }
       },
       "failures": []
@@ -1560,44 +1560,44 @@
         "path": "repo:docs/data/scope_claims_audit.json",
         "exists": true,
         "bytes": 21630,
-        "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+        "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         },
         "hf_model_data": {
           "path": "hf_model:data/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         },
         "hf_model": {
           "path": "hf_model:metrics/scope_claims_audit.json",
           "exists": true,
           "bytes": 21630,
-          "sha256": "6a80b6012bbfba66cba524432b08421496ef0ba1d6eee6c7e5db8b382aa0f3ba"
+          "sha256": "947dfce64ff88a0f2dd7b1f72091aa61ccd58596525e2e43fd63c9b02ef23f62"
         }
       },
       "failures": []
@@ -1658,44 +1658,44 @@
         "path": "repo:docs/data/source_alignment_audit.json",
         "exists": true,
         "bytes": 4432,
-        "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+        "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         },
         "hf_model_data": {
           "path": "hf_model:data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         },
         "hf_model": {
           "path": "hf_model:metrics/source_alignment_audit.json",
           "exists": true,
           "bytes": 4432,
-          "sha256": "98bffea722a8d50936e214538b37766d3952ca362854757bf31121031fdecbc8"
+          "sha256": "ba571e397b5dfe71f82efc83dc30a44ec693b92f1316e6dd52adca12811ad52f"
         }
       },
       "failures": []
@@ -1756,44 +1756,44 @@
         "path": "repo:docs/data/single_episode_task_model_radar.json",
         "exists": true,
         "bytes": 51064,
-        "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+        "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         },
         "hf_model_data": {
           "path": "hf_model:data/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         },
         "hf_model": {
           "path": "hf_model:metrics/single_episode_task_model_radar.json",
           "exists": true,
           "bytes": 51064,
-          "sha256": "5176552d9b735d36a1033277ad64b68038cc27605565ec9d714397d2a6de7623"
+          "sha256": "dad53d1c12eb8947b064372e5ccd7c416cefab6e57175cd990e60af9e56ec958"
         }
       },
       "failures": []
@@ -1804,45 +1804,45 @@
       "local": {
         "path": "repo:docs/data/episode128_task_model_radar.json",
         "exists": true,
-        "bytes": 184785,
-        "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+        "bytes": 184569,
+        "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         },
         "hf_model_data": {
           "path": "hf_model:data/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         },
         "hf_model": {
           "path": "hf_model:metrics/episode128_task_model_radar.json",
           "exists": true,
-          "bytes": 184785,
-          "sha256": "5d6119e3233992b9d4b4d9d9dc9a42024b419bb95de78028f9831e87fac6e88e"
+          "bytes": 184569,
+          "sha256": "a07df912c0909bd6067eb81a0f54856c1ed6da5ca5feee54f6b32d3296b3724e"
         }
       },
       "failures": []
@@ -2001,44 +2001,44 @@
         "path": "repo:docs/data/task_surface_integrity.json",
         "exists": true,
         "bytes": 45779,
-        "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+        "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         },
         "hf_model_data": {
           "path": "hf_model:data/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         },
         "hf_model": {
           "path": "hf_model:metrics/task_surface_integrity.json",
           "exists": true,
           "bytes": 45779,
-          "sha256": "4a6ef81221417db753a9324daab4f5826fbd92d192a97680b7234cff5817c9f3"
+          "sha256": "47c81fe3645e87eefe015d7fac0c23fb5a5411efce37f7d2b61ccc5398b03b1d"
         }
       },
       "failures": []
@@ -2098,45 +2098,45 @@
       "local": {
         "path": "repo:docs/data/task_method_20_result_matrix.json",
         "exists": true,
-        "bytes": 128510,
-        "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+        "bytes": 128400,
+        "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         },
         "hf_model_data": {
           "path": "hf_model:data/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         },
         "hf_model": {
           "path": "hf_model:metrics/task_method_20_result_matrix.json",
           "exists": true,
-          "bytes": 128510,
-          "sha256": "13722fe288669b640230ac894909d247f348cf0b3477f6f223943b87c728af1a"
+          "bytes": 128400,
+          "sha256": "afee9a09eeb7406a8f2ef4eb4205697fb9e05e5bad132686f21d28f93ac204a2"
         }
       },
       "failures": []
@@ -2147,45 +2147,45 @@
       "local": {
         "path": "repo:docs/data/task_method_20_gap_audit.json",
         "exists": true,
-        "bytes": 20037,
-        "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+        "bytes": 16980,
+        "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         },
         "hf_model_data": {
           "path": "hf_model:data/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         },
         "hf_model": {
           "path": "hf_model:metrics/task_method_20_gap_audit.json",
           "exists": true,
-          "bytes": 20037,
-          "sha256": "1e96af5557f4617e160233eb7d2180644f29fa9a8c67afed710ac0425fc71ddd"
+          "bytes": 16980,
+          "sha256": "50d966cb219b1bf89efda1bc0d1a0ed25ee058c65ca981cec37e67ea3e5f8061"
         }
       },
       "failures": []
@@ -2245,45 +2245,45 @@
       "local": {
         "path": "repo:docs/data/unified_task_model_radar.json",
         "exists": true,
-        "bytes": 228639,
-        "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+        "bytes": 228423,
+        "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         },
         "hf_model_data": {
           "path": "hf_model:data/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         },
         "hf_model": {
           "path": "hf_model:metrics/unified_task_model_radar.json",
           "exists": true,
-          "bytes": 228639,
-          "sha256": "5e163cc77679766fd86fdc40b324e61417f5a355b447922c2fce1d17ba52730e"
+          "bytes": 228423,
+          "sha256": "a84269541f448eb6b9d2ad29199dc2cab23c55ad4eacb1f0b4fbd2948e1b75c4"
         }
       },
       "failures": []
@@ -2295,44 +2295,44 @@
         "path": "repo:docs/data/website_integrity.json",
         "exists": true,
         "bytes": 20023,
-        "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+        "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:data/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         },
         "hf_artifacts_data": {
           "path": "hf_artifacts:data/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         },
         "hf_model_data": {
           "path": "hf_model:data/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         },
         "hf_model_docs_data": {
           "path": "hf_model:docs/data/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         },
         "hf_model": {
           "path": "hf_model:metrics/website_integrity.json",
           "exists": true,
           "bytes": 20023,
-          "sha256": "fb7171fd1d4962c3b8b289a1499c2eacb5404bf25587df34043c8712ef803acc"
+          "sha256": "2e525e86c7ecc338a5c5578eb6b8b1147e2b69a8ed632c942ae93c42d4e93876"
         }
       },
       "failures": []
@@ -2515,33 +2515,33 @@
       "local": {
         "path": "repo:docs/assets/charts/episode128_task_model_radar.svg",
         "exists": true,
-        "bytes": 50154,
-        "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+        "bytes": 50599,
+        "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:assets/charts/episode128_task_model_radar.svg",
           "exists": true,
-          "bytes": 50154,
-          "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+          "bytes": 50599,
+          "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/assets/charts/episode128_task_model_radar.svg",
           "exists": true,
-          "bytes": 50154,
-          "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+          "bytes": 50599,
+          "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
         },
         "hf_artifacts_card": {
           "path": "hf_artifacts:assets/charts/episode128_task_model_radar.svg",
           "exists": true,
-          "bytes": 50154,
-          "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+          "bytes": 50599,
+          "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
         },
         "hf_model": {
           "path": "hf_model:assets/charts/episode128_task_model_radar.svg",
           "exists": true,
-          "bytes": 50154,
-          "sha256": "be065f88f043241ce9d006bdc4a4e90b3b44ce5e04a8370ce65ce9b8f66fa5bc"
+          "bytes": 50599,
+          "sha256": "ec9cbf2a69d5b5cd466500dfbeaf996fa666af94a47a597abcc2465bbfad0bad"
         }
       },
       "failures": []
@@ -2589,33 +2589,33 @@
       "local": {
         "path": "repo:docs/assets/charts/unified_task_model_radar.svg",
         "exists": true,
-        "bytes": 56167,
-        "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+        "bytes": 56612,
+        "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:assets/charts/unified_task_model_radar.svg",
           "exists": true,
-          "bytes": 56167,
-          "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+          "bytes": 56612,
+          "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/assets/charts/unified_task_model_radar.svg",
           "exists": true,
-          "bytes": 56167,
-          "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+          "bytes": 56612,
+          "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
         },
         "hf_artifacts_card": {
           "path": "hf_artifacts:assets/charts/unified_task_model_radar.svg",
           "exists": true,
-          "bytes": 56167,
-          "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+          "bytes": 56612,
+          "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
         },
         "hf_model": {
           "path": "hf_model:assets/charts/unified_task_model_radar.svg",
           "exists": true,
-          "bytes": 56167,
-          "sha256": "1953a2c56b8c5a11a7def73c2899a49dc6c20d0cef3ae63c53fb10bfec7b7264"
+          "bytes": 56612,
+          "sha256": "03a3eb9369cda807197220eaa22282dec49b6119fa24c211a67c520daf63a611"
         }
       },
       "failures": []
@@ -5638,39 +5638,39 @@
       "local": {
         "path": "repo:docs/index.html",
         "exists": true,
-        "bytes": 254760,
-        "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+        "bytes": 254759,
+        "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:index.html",
           "exists": true,
-          "bytes": 254760,
-          "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+          "bytes": 254759,
+          "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
         },
         "hf_artifacts_root": {
           "path": "hf_artifacts:index.html",
           "exists": true,
-          "bytes": 254760,
-          "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+          "bytes": 254759,
+          "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
         },
         "hf_artifacts_docs": {
           "path": "hf_artifacts:docs/index.html",
           "exists": true,
-          "bytes": 254760,
-          "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+          "bytes": 254759,
+          "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
         },
         "hf_model": {
           "path": "hf_model:index.html",
           "exists": true,
-          "bytes": 254760,
-          "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+          "bytes": 254759,
+          "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
         },
         "hf_model_docs": {
           "path": "hf_model:docs/index.html",
           "exists": true,
-          "bytes": 254760,
-          "sha256": "56300081529887298a0e412302866cb7539384f0ed73b16cd9240e605fd4046a"
+          "bytes": 254759,
+          "sha256": "f913334541de577289a05e8d8e2ad99b1e3f52c2ae889e7e484a97d9a24b5c46"
         }
       },
       "failures": []
@@ -12829,6 +12829,31 @@
       },
       "failures": []
     },
+    {
+      "name": "results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+        "exists": true,
+        "bytes": 10565,
+        "sha256": "8cb5a4b6f45680c9df0b6b313f3d86abe64940f5a99793cffaaed205f80175c5"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+          "exists": true,
+          "bytes": 10565,
+          "sha256": "8cb5a4b6f45680c9df0b6b313f3d86abe64940f5a99793cffaaed205f80175c5"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+          "exists": true,
+          "bytes": 10565,
+          "sha256": "8cb5a4b6f45680c9df0b6b313f3d86abe64940f5a99793cffaaed205f80175c5"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620.runner.log",
       "status": "pass",
@@ -19154,6 +19179,306 @@
       },
       "failures": []
     },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+        "exists": true,
+        "bytes": 496,
+        "sha256": "2ddec993b14e63c1b4d3164c55a3d496b96489262ddc950a4a3903158aa33c47"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+          "exists": true,
+          "bytes": 496,
+          "sha256": "2ddec993b14e63c1b4d3164c55a3d496b96489262ddc950a4a3903158aa33c47"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+          "exists": true,
+          "bytes": 496,
+          "sha256": "2ddec993b14e63c1b4d3164c55a3d496b96489262ddc950a4a3903158aa33c47"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+        "exists": true,
+        "bytes": 1606,
+        "sha256": "d0f5bc9ec3064b86775bfd33f0c4c6c40302cbee6f7809468db4315888e9f57c"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+          "exists": true,
+          "bytes": 1606,
+          "sha256": "d0f5bc9ec3064b86775bfd33f0c4c6c40302cbee6f7809468db4315888e9f57c"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+          "exists": true,
+          "bytes": 1606,
+          "sha256": "d0f5bc9ec3064b86775bfd33f0c4c6c40302cbee6f7809468db4315888e9f57c"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+        "exists": true,
+        "bytes": 1619,
+        "sha256": "3cb0ed942c5af21223c582dd0dc7467a326306f10d2f03592d120fa8dcb173c5"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+          "exists": true,
+          "bytes": 1619,
+          "sha256": "3cb0ed942c5af21223c582dd0dc7467a326306f10d2f03592d120fa8dcb173c5"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+          "exists": true,
+          "bytes": 1619,
+          "sha256": "3cb0ed942c5af21223c582dd0dc7467a326306f10d2f03592d120fa8dcb173c5"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+        "exists": true,
+        "bytes": 5774,
+        "sha256": "e1ed24bc3a6507b427bc390f1dbea83c09195ac7bb80b99ed87ac7cc2e90a24d"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+          "exists": true,
+          "bytes": 5774,
+          "sha256": "e1ed24bc3a6507b427bc390f1dbea83c09195ac7bb80b99ed87ac7cc2e90a24d"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+          "exists": true,
+          "bytes": 5774,
+          "sha256": "e1ed24bc3a6507b427bc390f1dbea83c09195ac7bb80b99ed87ac7cc2e90a24d"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+        "exists": true,
+        "bytes": 1565,
+        "sha256": "ebbf5bfef95d8093fcbc094d3a03f9d765bbee5f8d80da027e948c55c35d0026"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+          "exists": true,
+          "bytes": 1565,
+          "sha256": "ebbf5bfef95d8093fcbc094d3a03f9d765bbee5f8d80da027e948c55c35d0026"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+          "exists": true,
+          "bytes": 1565,
+          "sha256": "ebbf5bfef95d8093fcbc094d3a03f9d765bbee5f8d80da027e948c55c35d0026"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+        "exists": true,
+        "bytes": 858,
+        "sha256": "ebcc6d877fd412ba5d2383ac343672e65cbd5023deff9ebb51a38b64f0b0ea77"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+          "exists": true,
+          "bytes": 858,
+          "sha256": "ebcc6d877fd412ba5d2383ac343672e65cbd5023deff9ebb51a38b64f0b0ea77"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+          "exists": true,
+          "bytes": 858,
+          "sha256": "ebcc6d877fd412ba5d2383ac343672e65cbd5023deff9ebb51a38b64f0b0ea77"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+        "exists": true,
+        "bytes": 2130,
+        "sha256": "f46a66c4fd06e944129b052be009d23e243567fdc54c9e101852c11dfabb0637"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+          "exists": true,
+          "bytes": 2130,
+          "sha256": "f46a66c4fd06e944129b052be009d23e243567fdc54c9e101852c11dfabb0637"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+          "exists": true,
+          "bytes": 2130,
+          "sha256": "f46a66c4fd06e944129b052be009d23e243567fdc54c9e101852c11dfabb0637"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+        "exists": true,
+        "bytes": 1562,
+        "sha256": "a2cc896d36516569829b955d9cf2f5d8586804b1805b778ef4face1be9074963"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+          "exists": true,
+          "bytes": 1562,
+          "sha256": "a2cc896d36516569829b955d9cf2f5d8586804b1805b778ef4face1be9074963"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+          "exists": true,
+          "bytes": 1562,
+          "sha256": "a2cc896d36516569829b955d9cf2f5d8586804b1805b778ef4face1be9074963"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+        "exists": true,
+        "bytes": 1100734,
+        "sha256": "1fd8f293af1432d49653dd93f00924d07ba6117129e1fc8c6cdb3b7c8a2bfb05"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+          "exists": true,
+          "bytes": 1100734,
+          "sha256": "1fd8f293af1432d49653dd93f00924d07ba6117129e1fc8c6cdb3b7c8a2bfb05"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+          "exists": true,
+          "bytes": 1100734,
+          "sha256": "1fd8f293af1432d49653dd93f00924d07ba6117129e1fc8c6cdb3b7c8a2bfb05"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+        "exists": true,
+        "bytes": 1082556,
+        "sha256": "43c2c76dffb1661c2dfb9947b75ac29fc68c311b793648801f75a2b3a1592922"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+          "exists": true,
+          "bytes": 1082556,
+          "sha256": "43c2c76dffb1661c2dfb9947b75ac29fc68c311b793648801f75a2b3a1592922"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+          "exists": true,
+          "bytes": 1082556,
+          "sha256": "43c2c76dffb1661c2dfb9947b75ac29fc68c311b793648801f75a2b3a1592922"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+        "exists": true,
+        "bytes": 1077207,
+        "sha256": "6e163ccfd831801dd5dcb6d6336c08ce6be0099fa427d0291af97a846430b2a2"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+          "exists": true,
+          "bytes": 1077207,
+          "sha256": "6e163ccfd831801dd5dcb6d6336c08ce6be0099fa427d0291af97a846430b2a2"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+          "exists": true,
+          "bytes": 1077207,
+          "sha256": "6e163ccfd831801dd5dcb6d6336c08ce6be0099fa427d0291af97a846430b2a2"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+        "exists": true,
+        "bytes": 1072835,
+        "sha256": "af208b8afe02886022915f634927b073646fa15465f157d4e1f7cad3c3408f0b"
+      },
+      "mirrors": {
+        "hf_artifacts": {
+          "path": "hf_artifacts:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+          "exists": true,
+          "bytes": 1072835,
+          "sha256": "af208b8afe02886022915f634927b073646fa15465f157d4e1f7cad3c3408f0b"
+        },
+        "hf_model": {
+          "path": "hf_model:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+          "exists": true,
+          "bytes": 1072835,
+          "sha256": "af208b8afe02886022915f634927b073646fa15465f157d4e1f7cad3c3408f0b"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/RUN_REPORT.md",
       "status": "pass",
@@ -23267,6 +23592,25 @@
       },
       "failures": []
     },
+    {
+      "name": "space_results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+        "exists": true,
+        "bytes": 10565,
+        "sha256": "8cb5a4b6f45680c9df0b6b313f3d86abe64940f5a99793cffaaed205f80175c5"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620.launcher.log",
+          "exists": true,
+          "bytes": 10565,
+          "sha256": "8cb5a4b6f45680c9df0b6b313f3d86abe64940f5a99793cffaaed205f80175c5"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "space_results/omni_finetune/deferred_launchers/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620.runner.log",
       "status": "pass",
@@ -24445,6 +24789,234 @@
       },
       "failures": []
     },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+        "exists": true,
+        "bytes": 496,
+        "sha256": "2ddec993b14e63c1b4d3164c55a3d496b96489262ddc950a4a3903158aa33c47"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md",
+          "exists": true,
+          "bytes": 496,
+          "sha256": "2ddec993b14e63c1b4d3164c55a3d496b96489262ddc950a4a3903158aa33c47"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+        "exists": true,
+        "bytes": 1606,
+        "sha256": "d0f5bc9ec3064b86775bfd33f0c4c6c40302cbee6f7809468db4315888e9f57c"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/collection_validation.json",
+          "exists": true,
+          "bytes": 1606,
+          "sha256": "d0f5bc9ec3064b86775bfd33f0c4c6c40302cbee6f7809468db4315888e9f57c"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+        "exists": true,
+        "bytes": 1619,
+        "sha256": "3cb0ed942c5af21223c582dd0dc7467a326306f10d2f03592d120fa8dcb173c5"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json",
+          "exists": true,
+          "bytes": 1619,
+          "sha256": "3cb0ed942c5af21223c582dd0dc7467a326306f10d2f03592d120fa8dcb173c5"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+        "exists": true,
+        "bytes": 5774,
+        "sha256": "e1ed24bc3a6507b427bc390f1dbea83c09195ac7bb80b99ed87ac7cc2e90a24d"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json",
+          "exists": true,
+          "bytes": 5774,
+          "sha256": "e1ed24bc3a6507b427bc390f1dbea83c09195ac7bb80b99ed87ac7cc2e90a24d"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+        "exists": true,
+        "bytes": 1565,
+        "sha256": "ebbf5bfef95d8093fcbc094d3a03f9d765bbee5f8d80da027e948c55c35d0026"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json",
+          "exists": true,
+          "bytes": 1565,
+          "sha256": "ebbf5bfef95d8093fcbc094d3a03f9d765bbee5f8d80da027e948c55c35d0026"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+        "exists": true,
+        "bytes": 858,
+        "sha256": "ebcc6d877fd412ba5d2383ac343672e65cbd5023deff9ebb51a38b64f0b0ea77"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json",
+          "exists": true,
+          "bytes": 858,
+          "sha256": "ebcc6d877fd412ba5d2383ac343672e65cbd5023deff9ebb51a38b64f0b0ea77"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+        "exists": true,
+        "bytes": 2130,
+        "sha256": "f46a66c4fd06e944129b052be009d23e243567fdc54c9e101852c11dfabb0637"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/summary.json",
+          "exists": true,
+          "bytes": 2130,
+          "sha256": "f46a66c4fd06e944129b052be009d23e243567fdc54c9e101852c11dfabb0637"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+        "exists": true,
+        "bytes": 1562,
+        "sha256": "a2cc896d36516569829b955d9cf2f5d8586804b1805b778ef4face1be9074963"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json",
+          "exists": true,
+          "bytes": 1562,
+          "sha256": "a2cc896d36516569829b955d9cf2f5d8586804b1805b778ef4face1be9074963"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+        "exists": true,
+        "bytes": 1100734,
+        "sha256": "1fd8f293af1432d49653dd93f00924d07ba6117129e1fc8c6cdb3b7c8a2bfb05"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard0.progress.jsonl",
+          "exists": true,
+          "bytes": 1100734,
+          "sha256": "1fd8f293af1432d49653dd93f00924d07ba6117129e1fc8c6cdb3b7c8a2bfb05"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+        "exists": true,
+        "bytes": 1082556,
+        "sha256": "43c2c76dffb1661c2dfb9947b75ac29fc68c311b793648801f75a2b3a1592922"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard1.progress.jsonl",
+          "exists": true,
+          "bytes": 1082556,
+          "sha256": "43c2c76dffb1661c2dfb9947b75ac29fc68c311b793648801f75a2b3a1592922"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+        "exists": true,
+        "bytes": 1077207,
+        "sha256": "6e163ccfd831801dd5dcb6d6336c08ce6be0099fa427d0291af97a846430b2a2"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard2.progress.jsonl",
+          "exists": true,
+          "bytes": 1077207,
+          "sha256": "6e163ccfd831801dd5dcb6d6336c08ce6be0099fa427d0291af97a846430b2a2"
+        }
+      },
+      "failures": []
+    },
+    {
+      "name": "space_results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+      "status": "pass",
+      "local": {
+        "path": "repo:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+        "exists": true,
+        "bytes": 1072835,
+        "sha256": "af208b8afe02886022915f634927b073646fa15465f157d4e1f7cad3c3408f0b"
+      },
+      "mirrors": {
+        "hf_space": {
+          "path": "hf_space:results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620_shard3.progress.jsonl",
+          "exists": true,
+          "bytes": 1072835,
+          "sha256": "af208b8afe02886022915f634927b073646fa15465f157d4e1f7cad3c3408f0b"
+        }
+      },
+      "failures": []
+    },
     {
       "name": "space_results/omni_finetune/xperience10m_cosmos3_super_retrieval_task_probes_a100_textonly_prompatch_v2_20260620/RUN_REPORT.md",
       "status": "pass",
@@ -27539,27 +28111,27 @@
       "local": {
         "path": "repo:TASK_METHOD_20_GAP_AUDIT.md",
         "exists": true,
-        "bytes": 6398,
-        "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
+        "bytes": 5661,
+        "sha256": "58369ff12eae39ac2b1518623b63b6d070e633c7a88bb0a06e54c556f3027f06"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:TASK_METHOD_20_GAP_AUDIT.md",
           "exists": true,
-          "bytes": 6398,
-          "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
+          "bytes": 5661,
+          "sha256": "58369ff12eae39ac2b1518623b63b6d070e633c7a88bb0a06e54c556f3027f06"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:TASK_METHOD_20_GAP_AUDIT.md",
           "exists": true,
-          "bytes": 6398,
-          "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
+          "bytes": 5661,
+          "sha256": "58369ff12eae39ac2b1518623b63b6d070e633c7a88bb0a06e54c556f3027f06"
         },
         "hf_model": {
           "path": "hf_model:TASK_METHOD_20_GAP_AUDIT.md",
           "exists": true,
-          "bytes": 6398,
-          "sha256": "5d63a8ad52dcdd3624c9c73fd292121cf9a14644ad369784894c3fd1fcaa83a1"
+          "bytes": 5661,
+          "sha256": "58369ff12eae39ac2b1518623b63b6d070e633c7a88bb0a06e54c556f3027f06"
         }
       },
       "failures": []
@@ -27570,27 +28142,27 @@
       "local": {
         "path": "repo:TASK_METHOD_20_RESULT_MATRIX.md",
         "exists": true,
-        "bytes": 3771,
-        "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
+        "bytes": 3739,
+        "sha256": "325df952cc73a1a530781c3f55cf87105a178ac3cb3553515c817d00916deb27"
       },
       "mirrors": {
         "hf_space": {
           "path": "hf_space:TASK_METHOD_20_RESULT_MATRIX.md",
           "exists": true,
-          "bytes": 3771,
-          "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
+          "bytes": 3739,
+          "sha256": "325df952cc73a1a530781c3f55cf87105a178ac3cb3553515c817d00916deb27"
         },
         "hf_artifacts": {
           "path": "hf_artifacts:TASK_METHOD_20_RESULT_MATRIX.md",
           "exists": true,
-          "bytes": 3771,
-          "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
+          "bytes": 3739,
+          "sha256": "325df952cc73a1a530781c3f55cf87105a178ac3cb3553515c817d00916deb27"
         },
         "hf_model": {
           "path": "hf_model:TASK_METHOD_20_RESULT_MATRIX.md",
           "exists": true,
-          "bytes": 3771,
-          "sha256": "a17499bf2e6f81dbb5682abd42ae2f66ff2ae41d28ea8f4a437eb65350956fc6"
+          "bytes": 3739,
+          "sha256": "325df952cc73a1a530781c3f55cf87105a178ac3cb3553515c817d00916deb27"
         }
       },
       "failures": []

data/public_surface_qa.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Public Project Surface",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:51:35+00:00",
+  "generated_at_utc": "2026-06-20T15:22:58+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": "2026-06-20T14:49:37+00:00"
+          "generated_at_utc": "2026-06-20T15:22:30+00:00"
         },
         "rendered_site_check": {
           "exists": true,
@@ -28,27 +28,27 @@
         "task_surface_integrity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-20T14:04:25+00:00"
+          "generated_at_utc": "2026-06-20T14:51:35+00:00"
         },
         "source_alignment": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-20T14:03:48+00:00"
+          "generated_at_utc": "2026-06-20T15:22:29+00:00"
         },
         "scale_up_status": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-20T14:04:01+00:00"
+          "generated_at_utc": "2026-06-20T15:22:32+00:00"
         },
         "publication_package": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-20T14:50:09+00:00"
+          "generated_at_utc": "2026-06-20T15:22:46+00:00"
         },
         "mirror_parity": {
           "exists": true,
           "status": "pass",
-          "generated_at_utc": "2026-06-20T14:51:20+00:00"
+          "generated_at_utc": "2026-06-20T14:52:37+00:00"
         }
       },
       "failures": {}

data/quality_gates.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Release Checks",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:03:29+00:00",
+  "generated_at_utc": "2026-06-20T15:20:32+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/source_alignment_audit.json

@@ -1,7 +1,7 @@
 {
   "title": "Ropedia Xperience-10M Source Alignment Note",
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:51:35+00:00",
+  "generated_at_utc": "2026-06-20T15:22:29+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": "2026-06-20T13:58:05+00:00",
+  "generated_at_utc": "2026-06-20T15:21:33+00:00",
   "immediate_actions": [
     {
       "artifact": "docs/data/task_method_20_gap_audit.json",
       "id": "gap_audit",
-      "purpose": "Keep the 16 scoreless cells visible and reproducible."
+      "purpose": "Keep the 12 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": 15,
-      "scoreless_task_count": 5,
+      "scored_task_count": 19,
+      "scoreless_task_count": 1,
       "status_counts": {
-        "not_evaluated_in_verified_package": 5,
-        "scored": 15
+        "not_evaluated_in_verified_package": 1,
+        "scored": 19
       }
     },
     "metadata128_neural_mlp": {
@@ -135,12 +135,12 @@
   },
   "missing_by_method": {
     "cosmos3_nano_future_window": 9,
-    "cosmos3_super_reasoner": 5,
+    "cosmos3_super_reasoner": 1,
     "metadata128_neural_mlp": 1,
     "metadata128_simple": 1
   },
   "missing_by_status": {
-    "not_evaluated_in_verified_package": 14,
+    "not_evaluated_in_verified_package": 10,
     "not_supported_by_metadata_only_package": 1,
     "unsupported_without_required_target": 1
   },
@@ -158,23 +158,15 @@
       "cosmos3_nano_future_window"
     ],
     "11 Temporal Order Verification": [
-      "cosmos3_nano_future_window",
-      "cosmos3_super_reasoner"
+      "cosmos3_nano_future_window"
     ],
     "12 Multimodal Synchronization Detection": [
-      "cosmos3_nano_future_window",
-      "cosmos3_super_reasoner"
-    ],
-    "14 Long-Horizon Next-Subtask Forecasting": [
-      "cosmos3_super_reasoner"
+      "cosmos3_nano_future_window"
     ],
     "15 Interaction Text Prediction": [
       "cosmos3_nano_future_window",
       "cosmos3_super_reasoner"
     ],
-    "17 Future Object-Set Forecasting": [
-      "cosmos3_super_reasoner"
-    ],
     "18 IMU-to-Hand Pose Reconstruction": [
       "cosmos3_nano_future_window"
     ],
@@ -237,19 +229,6 @@
       "task_label": "Language Grounding",
       "task_number": 8
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "f1",
-      "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": "temporal_order",
-      "task_label": "Temporal Order Verification",
-      "task_number": 11
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "f1",
@@ -263,19 +242,6 @@
       "task_label": "Temporal Order Verification",
       "task_number": 11
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "f1",
-      "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": "misalignment_detection",
-      "task_label": "Multimodal Synchronization Detection",
-      "task_number": 12
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "f1",
@@ -289,19 +255,6 @@
       "task_label": "Multimodal Synchronization Detection",
       "task_number": 12
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "macro_f1",
-      "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": "next_subtask_forecast",
-      "task_label": "Long-Horizon Next-Subtask Forecasting",
-      "task_number": 14
-    },
     {
       "method": "Cosmos3-Super Reasoner",
       "metric_key": "macro_f1",
@@ -328,19 +281,6 @@
       "task_label": "Interaction Text Prediction",
       "task_number": 15
     },
-    {
-      "method": "Cosmos3-Super Reasoner",
-      "metric_key": "micro_f1",
-      "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": "object_set_forecast",
-      "task_label": "Future Object-Set Forecasting",
-      "task_number": 17
-    },
     {
       "method": "Cosmos3-Nano Future Window",
       "metric_key": "mae",
@@ -440,8 +380,8 @@
     "method_count": 9,
     "method_task_record_count": 180,
     "proxy_scored_method_task_count": 4,
-    "scored_method_task_count": 164,
-    "scoreless_method_task_count": 16,
+    "scored_method_task_count": 168,
+    "scoreless_method_task_count": 12,
     "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": "2026-06-20T13:58:04+00:00",
+  "generated_at_utc": "2026-06-20T15:20:32+00:00",
   "task_count": 20,
   "method_count": 9,
   "method_task_record_count": 180,
-  "scored_method_task_count": 164,
+  "scored_method_task_count": 168,
   "series": [
     {
       "id": "minimal",
@@ -183,17 +183,17 @@
       "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": 15,
-      "covered_task_count": 15,
+      "scored_task_count": 19,
+      "covered_task_count": 19,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 5,
+      "scoreless_task_count": 1,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 5,
+      "not_evaluated_task_count": 1,
       "status_counts": {
-        "not_evaluated_in_verified_package": 5,
-        "scored": 15
+        "not_evaluated_in_verified_package": 1,
+        "scored": 19
       },
-      "coverage_fraction": 0.75,
+      "coverage_fraction": 0.95,
       "result_record_fraction": 1.0
     },
     {
@@ -1974,17 +1974,17 @@
       "task_label": "Temporal Order Verification",
       "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": "f1",
-      "source": null,
+      "raw": 0.6286317274823326,
+      "raw_text": "0.6286",
+      "normalized_score": 0.6286317274823326,
+      "metric_key": "temporal_order_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 11,
@@ -2136,17 +2136,17 @@
       "task_label": "Multimodal Synchronization Detection",
       "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": "f1",
-      "source": null,
+      "raw": 0.37271645981034185,
+      "raw_text": "0.3727",
+      "normalized_score": 0.37271645981034185,
+      "metric_key": "misalignment_detection_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 12,
@@ -2460,17 +2460,17 @@
       "task_label": "Long-Horizon Next-Subtask 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": "macro_f1",
-      "source": null,
+      "raw": 0.0,
+      "raw_text": "0.0000",
+      "normalized_score": 0.0,
+      "metric_key": "next_subtask_forecast_macro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 14,
@@ -2946,17 +2946,17 @@
       "task_label": "Future Object-Set 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": "micro_f1",
-      "source": null,
+      "raw": 0.0009279881217520415,
+      "raw_text": "0.0009",
+      "normalized_score": 0.0009279881217520415,
+      "metric_key": "object_set_forecast_micro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 17,

data/task_surface_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:51:35+00:00",
+  "generated_at_utc": "2026-06-20T15:22:58+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": "2026-06-20T13:58:04+00:00",
+  "generated_at_utc": "2026-06-20T15:20:32+00:00",
   "task_count": 20,
   "method_count": 9,
   "method_task_record_count": 180,
-  "scored_method_task_count": 164,
+  "scored_method_task_count": 168,
   "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",
@@ -192,17 +192,17 @@
       "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": 15,
-      "covered_task_count": 15,
+      "scored_task_count": 19,
+      "covered_task_count": 19,
       "proxy_scored_task_count": 0,
-      "scoreless_task_count": 5,
+      "scoreless_task_count": 1,
       "unsupported_task_count": 0,
-      "not_evaluated_task_count": 5,
+      "not_evaluated_task_count": 1,
       "status_counts": {
-        "not_evaluated_in_verified_package": 5,
-        "scored": 15
+        "not_evaluated_in_verified_package": 1,
+        "scored": 19
       },
-      "coverage_fraction": 0.75,
+      "coverage_fraction": 0.95,
       "result_record_fraction": 1.0
     },
     {
@@ -1384,6 +1384,17 @@
           "raw_text": "0.4098",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.6286317274823326,
+          "metric_key": "temporal_order_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.6286317274823326,
+          "raw_text": "0.6286",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.4198864140782312,
           "metric_key": "f1",
@@ -1428,17 +1439,6 @@
           "raw_text": "0.8030",
           "status_label": "scored"
         },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "f1",
-          "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": "f1",
@@ -1495,6 +1495,17 @@
           "raw_text": "0.3345",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.37271645981034185,
+          "metric_key": "misalignment_detection_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.37271645981034185,
+          "raw_text": "0.3727",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.49980060227663614,
           "metric_key": "f1",
@@ -1539,17 +1550,6 @@
           "raw_text": "0.8273",
           "status_label": "scored"
         },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "f1",
-          "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": "f1",
@@ -1728,6 +1728,17 @@
           "raw_text": "0.0042",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.0,
+          "metric_key": "next_subtask_forecast_macro_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.0,
+          "raw_text": "0.0000",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.0001206030150753769,
           "metric_key": "macro_f1",
@@ -1771,17 +1782,6 @@
           "normalized_score": 0.0,
           "raw_text": "0.0000",
           "status_label": "scored"
-        },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "macro_f1",
-          "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"
         }
       }
     },
@@ -2061,6 +2061,17 @@
           "raw_text": "0.1659",
           "status_label": "scored"
         },
+        "cosmos3_super_reasoner": {
+          "raw": 0.0009279881217520415,
+          "metric_key": "object_set_forecast_micro_f1",
+          "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/metrics.json",
+          "scope": "multi_episode_128_partial_model_overlay",
+          "status": "scored",
+          "reason": null,
+          "normalized_score": 0.0009279881217520415,
+          "raw_text": "0.0009",
+          "status_label": "scored"
+        },
         "metadata128_simple": {
           "raw": 0.17656983343047333,
           "metric_key": "micro_f1",
@@ -2104,17 +2115,6 @@
           "normalized_score": 0.17523098630012288,
           "raw_text": "0.1752",
           "status_label": "scored"
-        },
-        "cosmos3_super_reasoner": {
-          "raw": null,
-          "metric_key": "micro_f1",
-          "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"
         }
       }
     },
@@ -2498,7 +2498,7 @@
       "id": "cosmos3_super_reasoner",
       "title": "Cosmos3-Super Reasoner",
       "status": "verified_base_weight_eval",
-      "coverage": "20 records / 15 scored task-aligned axes",
+      "coverage": "20 records / 19 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"
     },
@@ -4272,17 +4272,17 @@
       "task_label": "Temporal Order Verification",
       "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": "f1",
-      "source": null,
+      "raw": 0.6286317274823326,
+      "raw_text": "0.6286",
+      "normalized_score": 0.6286317274823326,
+      "metric_key": "temporal_order_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 11,
@@ -4434,17 +4434,17 @@
       "task_label": "Multimodal Synchronization Detection",
       "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": "f1",
-      "source": null,
+      "raw": 0.37271645981034185,
+      "raw_text": "0.3727",
+      "normalized_score": 0.37271645981034185,
+      "metric_key": "misalignment_detection_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 12,
@@ -4758,17 +4758,17 @@
       "task_label": "Long-Horizon Next-Subtask 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": "macro_f1",
-      "source": null,
+      "raw": 0.0,
+      "raw_text": "0.0000",
+      "normalized_score": 0.0,
+      "metric_key": "next_subtask_forecast_macro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 14,
@@ -5244,17 +5244,17 @@
       "task_label": "Future Object-Set 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": "micro_f1",
-      "source": null,
+      "raw": 0.0009279881217520415,
+      "raw_text": "0.0009",
+      "normalized_score": 0.0009279881217520415,
+      "metric_key": "object_set_forecast_micro_f1",
+      "source": "results/omni_finetune/verified_public/xperience10m_cosmos3_super_reasoner_128ep_test_full_20260607/eval/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": 17,

data/website_integrity.json

@@ -1,6 +1,6 @@
 {
   "status": "pass",
-  "generated_at_utc": "2026-06-20T14:49:37+00:00",
+  "generated_at_utc": "2026-06-20T15:22:30+00:00",
   "docs_root": "docs",
   "site_base": "/ropedia-xperience-10m-task-suite/",
   "summary": {
@@ -80,8 +80,8 @@
       "name": "project_overview_precedes_progress_ledger",
       "status": "pass",
       "reason": "The project overview should appear before the deeper progress ledger.",
-      "overview_index": 95752,
-      "evidence_index": 132135
+      "overview_index": 95751,
+      "evidence_index": 132134
     },
     {
       "name": "project_status_links_json",
@@ -159,9 +159,9 @@
       "name": "evaluation_protocol_between_overview_and_progress",
       "status": "pass",
       "reason": "The evaluation protocol should appear before the deeper evidence ledger.",
-      "overview_index": 95752,
-      "protocol_index": 128316,
-      "evidence_index": 132135
+      "overview_index": 95751,
+      "protocol_index": 128315,
+      "evidence_index": 132134
     },
     {
       "name": "evaluation_protocol_links_json",
@@ -301,7 +301,7 @@
     },
     {
       "path": "data/artifact_index.json",
-      "bytes": 116644,
+      "bytes": 116645,
       "top_level_type": "dict"
     },
     {
@@ -316,7 +316,7 @@
     },
     {
       "path": "data/episode128_task_model_radar.json",
-      "bytes": 184785,
+      "bytes": 184569,
       "top_level_type": "dict"
     },
     {
@@ -346,7 +346,7 @@
     },
     {
       "path": "data/live_publication_status.json",
-      "bytes": 181998,
+      "bytes": 181991,
       "top_level_type": "dict"
     },
     {
@@ -486,12 +486,12 @@
     },
     {
       "path": "data/task_method_20_gap_audit.json",
-      "bytes": 20037,
+      "bytes": 16980,
       "top_level_type": "dict"
     },
     {
       "path": "data/task_method_20_result_matrix.json",
-      "bytes": 128510,
+      "bytes": 128400,
       "top_level_type": "dict"
     },
     {
@@ -526,12 +526,12 @@
     },
     {
       "path": "data/unified_task_model_radar.json",
-      "bytes": 228639,
+      "bytes": 228423,
       "top_level_type": "dict"
     },
     {
       "path": "data/website_integrity.json",
-      "bytes": 20022,
+      "bytes": 20023,
       "top_level_type": "dict"
     },
     {
@@ -571,7 +571,7 @@
     {
       "path": "assets/charts/episode128_task_model_radar.svg",
       "exists": true,
-      "bytes": 50154,
+      "bytes": 50599,
       "format": "SVG",
       "has_viewbox": true
     },
@@ -641,7 +641,7 @@
     {
       "path": "assets/charts/unified_task_model_radar.svg",
       "exists": true,
-      "bytes": 56167,
+      "bytes": 56612,
       "format": "SVG",
       "has_viewbox": true
     },

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/RUN_REPORT.md

@@ -0,0 +1,11 @@
+# Cosmos3-Super Reasoner Future Task Probes
+
+- Run ID: `xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620`
+- Shards: `4`
+
+| Task | Metric | Score | Samples |
+| --- | --- | ---: | ---: |
+| Temporal Order Verification | temporal_order_f1 | 0.628632 | 3951 |
+| Multimodal Misalignment Detection | misalignment_detection_f1 | 0.372716 | 3951 |
+| Long-Horizon Next-Subtask Forecasting | macro_f1 | 0.000000 | 3951 |
+| Future Object-Set Forecasting | micro_f1 | 0.000928 | 3951 |

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/misalignment_detection/metrics.json

@@ -0,0 +1,33 @@
+{
+  "accuracy": 0.4986079473551,
+  "adapter_dir": ".",
+  "base_url": "http://127.0.0.1:8000/v1",
+  "dataset_jsonl": "results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl",
+  "eval_split": "test",
+  "future_frames": 100,
+  "labels": [
+    "aligned",
+    "shifted"
+  ],
+  "macro_f1": 0.37271645981034185,
+  "media_mode": "text_only",
+  "metric_key": "misalignment_detection_f1",
+  "misalignment_detection_accuracy": 0.4986079473551,
+  "misalignment_detection_f1": 0.37271645981034185,
+  "misalignment_detection_macro_f1": 0.37271645981034185,
+  "model": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "model_id": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "num_samples": 3951,
+  "primary_metric": "misalignment_detection_f1",
+  "primary_score": 0.37271645981034185,
+  "run_id": "xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620",
+  "sample_offset": 0,
+  "sample_stride": 1,
+  "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.",
+  "status": "pass",
+  "task_id": "misalignment_detection",
+  "task_label": "Multimodal Misalignment Detection",
+  "task_number": 12,
+  "title": "Cosmos3-Super Reasoner Multimodal Misalignment Detection"
+}

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/next_subtask_forecast/metrics.json

@@ -0,0 +1,157 @@
+{
+  "accuracy": 0.0,
+  "adapter_dir": ".",
+  "base_url": "http://127.0.0.1:8000/v1",
+  "dataset_jsonl": "results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl",
+  "eval_split": "test",
+  "future_frames": 100,
+  "labels": [
+    "Adjust lantern string and handle components",
+    "Adjusting and securing paper structure",
+    "Adjusting cookware",
+    "Adjusting paper edge and placing strip",
+    "Adjusting stock and finishing placement",
+    "Align paper lantern edges",
+    "Aligning canned goods on the shelf",
+    "Approach inventory boxes",
+    "Approaching the stove",
+    "Arrange buttons",
+    "Arrange buttons in a line",
+    "Assembling material pieces",
+    "Attempt to fit and place puzzle piece",
+    "Bending plastic strip",
+    "Browsing smartphone",
+    "Bundle display hooks",
+    "Checking cooking pot",
+    "Checking smartphone",
+    "Cleaning cloth maintenance",
+    "Cleaning kitchen surfaces",
+    "Cooking at the stove",
+    "Count beads and retrieve more",
+    "Cut along the marked line",
+    "Cut cardboard",
+    "Cutting cardboard pieces",
+    "Document bead counts",
+    "Entering the training area",
+    "Expand and adjust lantern shape",
+    "Expand paper lantern",
+    "Extract wire hangers from inventory",
+    "Final alignment of lantern edges",
+    "Finalizing shelf organization",
+    "Finish cutting along the marked line and reposition",
+    "Fold and grasp lantern",
+    "Folding plastic strip",
+    "Grasp lantern component",
+    "Greeting participants",
+    "Handle paper lantern component",
+    "Handling container lid",
+    "Handling earbud case",
+    "Handling plastic strip",
+    "Holding smartphone",
+    "Initiating assembly",
+    "Inspect shelf condition",
+    "Inspect shelf condition and observe surroundings",
+    "Install display hooks",
+    "Labeling cardboard pieces",
+    "Manipulate adhesive strip",
+    "Manipulate and release paper strips",
+    "Manipulate craft pieces",
+    "Manipulate paper decoration",
+    "Manipulate paper edge",
+    "Manipulate paper strip",
+    "Manipulate paper strips",
+    "Manipulate puzzle piece",
+    "Manipulate puzzle pieces",
+    "Manipulating beads",
+    "Manipulating plastic strip",
+    "Manipulating yellow strips",
+    "Move to stocking area",
+    "Moving around the kitchen",
+    "Moving through the room",
+    "Observe puzzle progress",
+    "Observing and pausing",
+    "Open paper lantern",
+    "Operate and release smartphone",
+    "Organize and count beads",
+    "Organizing cardboard pieces",
+    "Paper quilling craft",
+    "Pick up and place buttons",
+    "Pick up and place puzzle piece",
+    "Pick, place, and count beads",
+    "Picking up and placing canned goods",
+    "Place and adjust items on shelf",
+    "Place and retrieve items for stocking",
+    "Place items on shelf",
+    "Place puzzle piece",
+    "Placing and retrieving canned goods",
+    "Placing canned goods on the shelf",
+    "Prepare to cut cardboard",
+    "Preparing craft area",
+    "Preparing materials",
+    "Preparing plastic strip for folding",
+    "Reach for, pick up, and attempt to fit puzzle piece",
+    "Reaching into the box for more stock",
+    "Release and adjust puzzle piece",
+    "Release scissors and operate smartphone",
+    "Remove packaging and prepare component",
+    "Retrieving and placing canned goods",
+    "Retrieving cleaning supplies",
+    "Scrolling and placing smartphone down",
+    "Scrolling and setting down smartphone",
+    "Scrolling smartphone screen",
+    "Search for and pick up puzzle piece",
+    "Secure lantern with adhesive",
+    "Secure paper edges with adhesive",
+    "Setting up smartphone",
+    "Sort and adjust button line",
+    "Sort and arrange buttons",
+    "Sort and group beads",
+    "Sort and place buttons",
+    "Sort and record bead counts",
+    "Sort beads and record count",
+    "Sort buttons",
+    "Sort buttons by color",
+    "Sort craft materials into piles",
+    "Sort puzzle pieces",
+    "Sort small craft pieces",
+    "Stocking multiple cans on the shelf",
+    "Tidying workspace",
+    "Transitioning to smartphone usage",
+    "Unfold paper lantern",
+    "Unpack additional lantern component",
+    "Unpack and place items on shelf",
+    "Use smartphone",
+    "Using a smartphone",
+    "Using phone",
+    "Using phone and resuming work",
+    "Using smartphone",
+    "Walk to shelf location",
+    "Walking to the crafting area",
+    "Wiping counter",
+    "Working with paper strips",
+    "Working with paper strips and using phone",
+    "unknown",
+    "Adjusting and folding cardboard",
+    "Adjust and align Mahjong tiles"
+  ],
+  "macro_f1": 0.0,
+  "media_mode": "text_only",
+  "metric_key": "macro_f1",
+  "model": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "model_id": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "next_subtask_forecast_accuracy": 0.0,
+  "next_subtask_forecast_macro_f1": 0.0,
+  "num_samples": 3951,
+  "primary_metric": "macro_f1",
+  "primary_score": 0.0,
+  "run_id": "xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620",
+  "sample_offset": 0,
+  "sample_stride": 1,
+  "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.",
+  "status": "pass",
+  "task_id": "next_subtask_forecast",
+  "task_label": "Long-Horizon Next-Subtask Forecasting",
+  "task_number": 14,
+  "title": "Cosmos3-Super Reasoner Long-Horizon Next-Subtask Forecasting"
+}

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/object_set_forecast/metrics.json

@@ -0,0 +1,30 @@
+{
+  "adapter_dir": ".",
+  "base_url": "http://127.0.0.1:8000/v1",
+  "dataset_jsonl": "results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl",
+  "eval_split": "test",
+  "exact_match": 0.0002531004808909137,
+  "future_frames": 100,
+  "media_mode": "text_only",
+  "metric_key": "micro_f1",
+  "micro_f1": 0.0009279881217520415,
+  "model": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "model_id": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "num_samples": 3951,
+  "object_set_forecast_exact_match": 0.0002531004808909137,
+  "object_set_forecast_micro_f1": 0.0009279881217520415,
+  "precision": 0.0006463497398442297,
+  "primary_metric": "micro_f1",
+  "primary_score": 0.0009279881217520415,
+  "recall": 0.0016446015952635473,
+  "run_id": "xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620",
+  "sample_offset": 0,
+  "sample_stride": 1,
+  "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.",
+  "status": "pass",
+  "task_id": "object_set_forecast",
+  "task_label": "Future Object-Set Forecasting",
+  "task_number": 17,
+  "title": "Cosmos3-Super Reasoner Future Object-Set Forecasting"
+}

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/server_info.json

@@ -0,0 +1,30 @@
+{
+  "data": [
+    {
+      "created": 1781964084,
+      "id": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+      "max_model_len": 262144,
+      "object": "model",
+      "owned_by": "vllm",
+      "parent": null,
+      "permission": [
+        {
+          "allow_create_engine": false,
+          "allow_fine_tuning": false,
+          "allow_logprobs": true,
+          "allow_sampling": true,
+          "allow_search_indices": false,
+          "allow_view": true,
+          "created": 1781964084,
+          "group": null,
+          "id": "modelperm-aa95c0c70bbe5707",
+          "is_blocking": false,
+          "object": "model_permission",
+          "organization": "*"
+        }
+      ],
+      "root": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay"
+    }
+  ],
+  "object": "list"
+}

results/omni_finetune/xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620/temporal_order/metrics.json

@@ -0,0 +1,33 @@
+{
+  "accuracy": 0.6370539104024298,
+  "adapter_dir": ".",
+  "base_url": "http://127.0.0.1:8000/v1",
+  "dataset_jsonl": "results/omni_finetune/xperience10m_qwen3_omni_128ep_multiscale_cap96_v5_full8gpu_lora_dataset/dataset_a100_eval.jsonl",
+  "eval_split": "test",
+  "future_frames": 100,
+  "labels": [
+    "correct",
+    "reversed"
+  ],
+  "macro_f1": 0.6286317274823326,
+  "media_mode": "text_only",
+  "metric_key": "temporal_order_f1",
+  "model": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "model_id": "/mnt/kgc/chaoyue/ropedia-xperience10m/models/nvidia__Cosmos3-Super_reasoner_overlay",
+  "num_samples": 3951,
+  "primary_metric": "temporal_order_f1",
+  "primary_score": 0.6286317274823326,
+  "run_id": "xperience10m_cosmos3_super_future_task_probes_a100_textonly_v1_20260620",
+  "sample_offset": 0,
+  "sample_stride": 1,
+  "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.",
+  "status": "pass",
+  "task_id": "temporal_order",
+  "task_label": "Temporal Order Verification",
+  "task_number": 11,
+  "temporal_order_accuracy": 0.6370539104024298,
+  "temporal_order_f1": 0.6286317274823326,
+  "temporal_order_macro_f1": 0.6286317274823326,
+  "title": "Cosmos3-Super Reasoner Temporal Order Verification"
+}