diff --git a/.modelhub_state/recovery_intents.jsonl b/.modelhub_state/recovery_intents.jsonl
index d0eedb36..81f458db 100644
--- a/.modelhub_state/recovery_intents.jsonl
+++ b/.modelhub_state/recovery_intents.jsonl
@@ -677,7 +677,7 @@
 {"batchId": "f0c31a9d78274a05a8d63b12de111b9d", "completedAt": "2026-09-11T15:47:41.708955+00:00", "configFingerprint": "cbe5569cdb6c42553436b837dd5b8f0cfa03334642a56f4ff0137f952f8eb84b", "configSource": "modelhub_live", "createdAt": "2026-09-11T15:47:33.464433+00:00", "framework": "vllm_fix_tokenizer", "intentId": "3c1b51ef2edd4a648576ee4dd6eec096", "lastModified": "2026-09-09T06:50:48+00:00", "modelAddress": "https://modelscope.cn/models/ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "reason": null, "reconciledAt": "2026-09-11T16:11:15.095128+00:00", "repoId": "ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "safeConfigVector": {"gpuNum": 1, "maxModelLen": 4096}, "status": "recovered_active", "targetGpu": "Vastai_va16", "taskId": "4782467", "taskType": "text-generation"}
 {"batchId": "39789cb8c4c7440d8960964792a5409c", "completedAt": "2026-09-11T16:03:38.536681+00:00", "configFingerprint": "1566bbd6c69427ad32f7f86aeba4e39850a1a7af5c6f4ef3c9ca82f0e882f3c5", "configSource": "modelhub_live", "createdAt": "2026-09-11T16:03:10.715055+00:00", "framework": "vllm-mlu", "intentId": "2a09ba9958f54909820f93b615b90b5d", "lastModified": "2026-09-11T15:37:46+00:00", "modelAddress": "https://modelscope.cn/models/CohereLabs/tiny-aya-base-32K", "reason": null, "reconciledAt": "2026-09-11T16:11:15.095627+00:00", "repoId": "CohereLabs/tiny-aya-base-32K", "safeConfigVector": {"dtype": "-", "gpuNum": 1}, "status": "recovered_active", "targetGpu": "Cambricon_mlu-370-x8", "taskId": "4782663", "taskType": "text-generation"}
 {"batchId": "89957c40421d44c3befe11112510797d", "completedAt": "2026-09-11T16:05:50.594050+00:00", "configFingerprint": "36ec94e289a8da5fd70b98a2f401fc5159740daf9e2a076da4fd490f34bcbc0b", "configSource": "modelhub_live", "createdAt": "2026-09-11T16:05:30.591889+00:00", "framework": "vllm-mlu", "intentId": "cee68951f2094549b8b727f182d8fd93", "lastModified": "2026-09-09T06:50:48+00:00", "modelAddress": "https://modelscope.cn/models/ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "reason": null, "reconciledAt": "2026-09-11T16:11:15.096364+00:00", "repoId": "ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "safeConfigVector": {"dtype": "-", "gpuNum": 1}, "status": "recovered_active", "targetGpu": "Cambricon_mlu-370-x4", "taskId": "4782685", "taskType": "text-generation"}
-{"batchId": "2232581eab6c4ac8b1132771afa56d98", "configFingerprint": "36ec94e289a8da5fd70b98a2f401fc5159740daf9e2a076da4fd490f34bcbc0b", "configSource": "modelhub_live", "createdAt": "2026-09-11T16:20:37.629854+00:00", "framework": "vllm-mlu", "intentId": "6456b20a2a6141ce834b8262c44747a0", "lastModified": "2026-09-11T15:37:46+00:00", "modelAddress": "https://modelscope.cn/models/CohereLabs/tiny-aya-base-32K", "repoId": "CohereLabs/tiny-aya-base-32K", "safeConfigVector": {"dtype": "-", "gpuNum": 1}, "status": "pending", "targetGpu": "Cambricon_mlu-370-x4", "taskType": "text-generation"}
+{"batchId": "2232581eab6c4ac8b1132771afa56d98", "completedAt": "2026-09-11T16:20:46.708114+00:00", "configFingerprint": "36ec94e289a8da5fd70b98a2f401fc5159740daf9e2a076da4fd490f34bcbc0b", "configSource": "modelhub_live", "createdAt": "2026-09-11T16:20:37.629854+00:00", "framework": "vllm-mlu", "intentId": "6456b20a2a6141ce834b8262c44747a0", "lastModified": "2026-09-11T15:37:46+00:00", "modelAddress": "https://modelscope.cn/models/CohereLabs/tiny-aya-base-32K", "reason": null, "repoId": "CohereLabs/tiny-aya-base-32K", "safeConfigVector": {"dtype": "-", "gpuNum": 1}, "status": "submitted", "targetGpu": "Cambricon_mlu-370-x4", "taskId": "4782813", "taskType": "text-generation"}
 {"batchId": "3cfa37416a9349629e882947d2816a4f", "completedAt": "2026-09-11T10:27:40.997245+00:00", "configFingerprint": "1cc16a7df03cf7b1485e2fda7e5c05e286a546d04e80bdd595c015004b3852a8", "configSource": "modelhub_live", "createdAt": "2026-09-11T10:27:32.517524+00:00", "framework": "vllm_0_17_0_corex_4_4_0", "intentId": "11684395861643dd8cc4c1610b8518a9", "repoId": "nv-community/Qwen3.8-27B-NVFP4", "safeConfigVector": {"gpuNum": 1, "maxModelLen": 4096}, "status": "uniqueness_rejected", "targetGpu": "Iluvatar_bi-150", "taskId": null, "taskType": "text-generation"}
 {"batchId": "3cfa37416a9349629e882947d2816a4f", "completedAt": "2026-09-11T10:27:40.997200+00:00", "configFingerprint": "e57ef22c2d8b2b75e495765c383ba0c5b795fe18d7b7f1baf8dcfadd87065bd8", "configSource": "modelhub_live", "createdAt": "2026-09-11T10:27:32.516990+00:00", "framework": "llamacpp", "intentId": "a0656dc70ed9494580076f5910d1e348", "repoId": "TokenRhythm/NeoHorse-1-4B-GGUF", "safeConfigVector": {"gpuNum": 1, "maxModelLen": 4096}, "status": "uniqueness_rejected", "targetGpu": "Iluvatar_bi-150", "taskId": null, "taskType": "text-generation"}
 {"batchId": "75d7acb27d704801b2ce084de0bf3900", "completedAt": "2026-09-11T02:14:19.417302+00:00", "configFingerprint": "3487754ba6ce87b4107d75c6476a4c5e5528f9179ac18665ecda0531c247da5e", "configSource": "modelhub_live", "createdAt": "2026-09-11T02:14:10.788042+00:00", "framework": "llamacpp", "intentId": "a31ea00fe3a847e6b52279c71b1fd09a", "repoId": "TokenRhythm/NeoHorse-1-4B-GGUF", "safeConfigVector": {"gpuNum": 1, "maxModelLen": 4096}, "status": "uniqueness_rejected", "targetGpu": "Ascend_910-b4", "taskId": null, "taskType": "text-generation"}
diff --git a/.modelhub_state/routing_intelligence.json b/.modelhub_state/routing_intelligence.json
index db74617c..c210585c 100644
--- a/.modelhub_state/routing_intelligence.json
+++ b/.modelhub_state/routing_intelligence.json
@@ -1,6 +1,6 @@
 {
   "acceptedByCategory": {
-    "unified_success_first": 812
+    "unified_success_first": 813
   },
   "acceptedByRoute": {
     "text-generation|Ascend_910-b3|llamacpp": 11,
@@ -9,7 +9,7 @@
     "text-generation|Ascend_910-b4|llamacpp": 8,
     "text-generation|Ascend_910-b4|vllm_tokenizer_patch": 32,
     "text-generation|Biren_166m|vllm": 52,
-    "text-generation|Cambricon_mlu-370-x4|vllm-mlu": 29,
+    "text-generation|Cambricon_mlu-370-x4|vllm-mlu": 30,
     "text-generation|Cambricon_mlu-370-x8|vllm": 13,
     "text-generation|Cambricon_mlu-370-x8|vllm-customized": 1,
     "text-generation|Cambricon_mlu-370-x8|vllm-mlu": 68,
@@ -26,8 +26,8 @@
     "text-generation|hygon_k100-ai|llamacpp": 11,
     "text-generation|hygon_k100-ai|vllm-patch-tokenizer": 45
   },
-  "acceptedSinceRefresh": 812,
-  "acceptedTotal": 812,
-  "generatedAt": "2026-09-11T16:05:50.528556+00:00",
+  "acceptedSinceRefresh": 813,
+  "acceptedTotal": 813,
+  "generatedAt": "2026-09-11T16:20:46.678522+00:00",
   "version": 1
 }
diff --git a/ledger/submissions.jsonl b/ledger/submissions.jsonl
index 0a2d1afe..c19f97ed 100644
--- a/ledger/submissions.jsonl
+++ b/ledger/submissions.jsonl
@@ -769,3 +769,4 @@
 {"framework": "vllm_fix_tokenizer", "modelAddress": "https://modelscope.cn/models/JANGQ-AI/AppleScript-8B-JANG_4M", "modelId": "JANGQ-AI/AppleScript-8B-JANG_4M", "submitTime": "2026-09-08T15:58:20.200513+00:00", "targetGpu": "Vastai_va16", "taskId": "4722293", "taskType": "text-generation", "templateId": "modelhub-live-text-generation-vllm-fix-tokenizer-vastai-va16"}
 {"framework": "vllm_fix_tokenizer", "modelAddress": "https://modelscope.cn/models/OpenBMB/BitCPM-CANN-8B", "modelId": "OpenBMB/BitCPM-CANN-8B", "submitTime": "2026-09-09T12:30:12.314907+00:00", "targetGpu": "Vastai_va16", "taskId": "4740401", "taskType": "text-generation", "templateId": "modelhub-live-text-generation-vllm-fix-tokenizer-vastai-va16"}
 {"framework": "vllm_fix_tokenizer", "modelAddress": "https://modelscope.cn/models/Youssofal/Qwen3.8-27B-MTPLX-Bare-Speed-FP16", "modelId": "Youssofal/Qwen3.8-27B-MTPLX-Bare-Speed-FP16", "submitTime": "2026-09-09T21:08:22.097030+00:00", "targetGpu": "Kunlunxin_p-800", "taskId": "4746441", "taskType": "text-generation", "templateId": "modelhub-live-text-generation-vllm-fix-tokenizer-kunlunxin-p-800"}
+{"framework": "vllm-mlu", "modelAddress": "https://modelscope.cn/models/CohereLabs/tiny-aya-base-32K", "modelId": "CohereLabs/tiny-aya-base-32K", "submitTime": "2026-09-11T16:20:38.542601+00:00", "targetGpu": "Cambricon_mlu-370-x4", "taskId": "4782813", "taskType": "text-generation", "templateId": "modelhub-live-text-generation-vllm-mlu-cambricon-mlu-370-x4"}
diff --git a/manifest.json b/manifest.json
index 73612fc0..7ceb95b7 100644
--- a/manifest.json
+++ b/manifest.json
@@ -9,16 +9,16 @@
     ".modelhub_state/queue_cleanup_latest.json": "aef9d56345e6e6b2c6e9bcc5c224abe9dc0699f57e994e20e663ac3e21dd4b56",
     ".modelhub_state/recent_outcomes.jsonl": "120eccf9fa38b19d315a7cd6d5484fed6a29bc5c3a6301a857e192af6e539172",
     ".modelhub_state/recovery_active_tasks.jsonl": "bc33b037b004570d0bd139fb08408fb9ba2aeaf6c522ae0287129c972ce8fc7c",
-    ".modelhub_state/recovery_intents.jsonl": "9bff535d7cb2b834f5341e63f712b837e6e6453a22be3bafc8267ba0e47ef2c9",
-    ".modelhub_state/routing_intelligence.json": "db347db1b10096edb1d7e17f9f9f7bdcbfcb4822b52a057e45c41bec6c635473",
+    ".modelhub_state/recovery_intents.jsonl": "6c1c9f2eb771389944e1bab073feddb04cc120d7e37cd7b5167dcb791e04afc9",
+    ".modelhub_state/routing_intelligence.json": "bf8485403bcc8fa616d8e0490877f80b66bf88a80a9abba70f6efc368e8b9ca5",
     ".modelhub_state/submission_exclusions.jsonl": "3790b5c148f222a4de26d50a3d8afc58891c8da251a82d20dbaf4fc35e6dbd89",
     ".modelhub_state/worker_crashes.jsonl": "7fa483580e87374493226e80bcf40bf8044847acba70045c96fa175966a1a8ba",
-    "ledger/submissions.jsonl": "1f823346c8c26c084eae21aaaf1c1dcf53f49c80a906a9984d19c8bfb7523571",
-    "outcomes/submissions.jsonl": "c0c1f06f5a692166c55ed57f0551c29a4469e548bf1f67c555b4e5dd8762fad1"
+    "ledger/submissions.jsonl": "ba9cdbb00234013f7033dd274a9c8671440eecc66e48ac4022c6b9524d1c94b5",
+    "outcomes/submissions.jsonl": "f5daff1f33466ca43d729ff9ca1027873cf8d9a89c3a25ddea9f0c8735835013"
   },
-  "generation": 5772,
-  "phase": "intent",
+  "generation": 5773,
+  "phase": "result",
   "schemaVersion": 1,
-  "updatedAt": "2026-09-11T16:20:37.695325+00:00",
+  "updatedAt": "2026-09-11T16:20:46.762902+00:00",
   "writerId": "18252d8a8ef94333abd55823b2d81c42"
 }
diff --git a/outcomes/submissions.jsonl b/outcomes/submissions.jsonl
index 54af8203..17654d0d 100644
--- a/outcomes/submissions.jsonl
+++ b/outcomes/submissions.jsonl
@@ -609,3 +609,4 @@
 {"failReason": null, "framework": "vllm_fix_tokenizer", "lastSyncTime": null, "modelId": "ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "modelProfile": {"architectures": ["Qwen3_5ForConditionalGeneration"], "configFingerprint": "cbe5569cdb6c42553436b837dd5b8f0cfa03334642a56f4ff0137f952f8eb84b", "configOptimization": {"applied": false, "source": "official"}, "configSource": "modelhub_live", "estimatedLoadBytes": 12999919368, "estimatedRequiredGiB": 14.555, "gpuMemoryEvidence": {"memoryGiB": 32.0, "source": "local_modelhub_preflight_oom"}, "gpuMemoryGiB": 32.0, "maximumRepositorySizeGiB": 26.667, "memorySizingBasis": "recursive_repository_on_disk", "modelCard": {}, "modelType": "qwen3_5", "modelscopeFileSize": 13023155398, "modelscopeLicense": "apache-2.0", "modelscopeParams": 3544073216, "modelscopeTags": ["license:apache-2.0", "model_type:qwen3_5", "library:mlx", "library:safetensors", "library:pytorch", "task:text-generation", "custom_tag:mlx", "custom_tag:mlx-serve", "custom_tag:qwen3_5", "custom_tag:apple-silicon", "custom_tag:imatrix", "custom_tag:mtp", "custom_tag:speculative-decoding"], "modelscopeTasks": ["text-generation"], "quantizationMethod": null, "repositoryOnDiskBytes": 13023155398}, "outcome": "pending", "status": "pending", "submitTime": "2026-09-11T15:47:34.151696+00:00", "targetGpu": "Vastai_va16", "taskId": "4782467", "taskType": "text-generation", "verifyResult": null}
 {"failReason": null, "framework": "vllm-mlu", "lastSyncTime": null, "modelId": "CohereLabs/tiny-aya-base-32K", "modelProfile": {"architectures": ["Cohere2ForCausalLM"], "configFingerprint": "1566bbd6c69427ad32f7f86aeba4e39850a1a7af5c6f4ef3c9ca82f0e882f3c5", "configOptimization": {"applied": false, "source": "official"}, "configSource": "modelhub_live", "estimatedLoadBytes": 6700539432, "estimatedRequiredGiB": 7.522, "gpuMemoryEvidence": {"memoryGiB": 24.0, "source": "local_modelhub_preflight_oom"}, "gpuMemoryGiB": 24.0, "maximumRepositorySizeGiB": 20.0, "memorySizingBasis": "recursive_repository_on_disk", "modelCard": {}, "modelType": "cohere2", "modelscopeFileSize": 3443, "modelscopeLicense": "cc-by-nc-4.0", "modelscopeParams": 3350243328, "modelscopeTags": ["license:cc-by-nc-4.0", "model_type:cohere2", "library:transformer", "library:safetensors", "library:pytorch", "task:text-generation", "custom_tag:aya", "custom_tag:multilingual", "custom_tag:long-context", "custom_tag:tiny-aya"], "modelscopeTasks": ["text-generation"], "quantizationMethod": null, "repositoryOnDiskBytes": 6730839322}, "outcome": "pending", "status": "pending", "submitTime": "2026-09-11T16:03:11.891697+00:00", "targetGpu": "Cambricon_mlu-370-x8", "taskId": "4782663", "taskType": "text-generation", "verifyResult": null}
 {"failReason": null, "framework": "vllm-mlu", "lastSyncTime": null, "modelId": "ddalcu/Qwen3.8-27B-MLX-Serve-iQ-MLX-3.8bpw", "modelProfile": {"architectures": ["Qwen3_5ForConditionalGeneration"], "configFingerprint": "36ec94e289a8da5fd70b98a2f401fc5159740daf9e2a076da4fd490f34bcbc0b", "configOptimization": {"applied": false, "source": "official"}, "configSource": "modelhub_live", "estimatedLoadBytes": 12999919368, "estimatedRequiredGiB": 14.555, "gpuMemoryEvidence": {"memoryGiB": 24.0, "source": "local_modelhub_preflight_oom"}, "gpuMemoryGiB": 24.0, "maximumRepositorySizeGiB": 20.0, "memorySizingBasis": "recursive_repository_on_disk", "modelCard": {}, "modelType": "qwen3_5", "modelscopeFileSize": 13023155398, "modelscopeLicense": "apache-2.0", "modelscopeParams": 3544073216, "modelscopeTags": ["license:apache-2.0", "model_type:qwen3_5", "library:mlx", "library:safetensors", "library:pytorch", "task:text-generation", "custom_tag:mlx", "custom_tag:mlx-serve", "custom_tag:qwen3_5", "custom_tag:apple-silicon", "custom_tag:imatrix", "custom_tag:mtp", "custom_tag:speculative-decoding"], "modelscopeTasks": ["text-generation"], "quantizationMethod": null, "repositoryOnDiskBytes": 13023155398}, "outcome": "pending", "status": "pending", "submitTime": "2026-09-11T16:05:31.441883+00:00", "targetGpu": "Cambricon_mlu-370-x4", "taskId": "4782685", "taskType": "text-generation", "verifyResult": null}
+{"failReason": null, "framework": "vllm-mlu", "lastSyncTime": null, "modelId": "CohereLabs/tiny-aya-base-32K", "modelProfile": {"architectures": ["Cohere2ForCausalLM"], "configFingerprint": "36ec94e289a8da5fd70b98a2f401fc5159740daf9e2a076da4fd490f34bcbc0b", "configOptimization": {"applied": false, "source": "official"}, "configSource": "modelhub_live", "estimatedLoadBytes": 6700539432, "estimatedRequiredGiB": 7.522, "gpuMemoryEvidence": {"memoryGiB": 24.0, "source": "local_modelhub_preflight_oom"}, "gpuMemoryGiB": 24.0, "maximumRepositorySizeGiB": 20.0, "memorySizingBasis": "recursive_repository_on_disk", "modelCard": {}, "modelType": "cohere2", "modelscopeFileSize": 3443, "modelscopeLicense": "cc-by-nc-4.0", "modelscopeParams": 3350243328, "modelscopeTags": ["license:cc-by-nc-4.0", "model_type:cohere2", "library:transformer", "library:safetensors", "library:pytorch", "task:text-generation", "custom_tag:aya", "custom_tag:multilingual", "custom_tag:long-context", "custom_tag:tiny-aya"], "modelscopeTasks": ["text-generation"], "quantizationMethod": null, "repositoryOnDiskBytes": 6730839322}, "outcome": "pending", "status": "pending", "submitTime": "2026-09-11T16:20:38.542601+00:00", "targetGpu": "Cambricon_mlu-370-x4", "taskId": "4782813", "taskType": "text-generation", "verifyResult": null}