Publish MT11-HIGH-L45 v1.0.0 dynamic LoRA

BASE_MODEL_REFERENCE.json

@@ -0,0 +1,10 @@
+{
+  "schema": "mt11-base-model-reference/1",
+  "model_id": "Qwen/Qwen3-4B-Base",
+  "local_training_path": "/workspace/MATH-NOTHINK-AIME-P1-v1.0/model/mn8_eos_r1_2/Qwen3-4B-Base-906bfd4",
+  "revision": "906bfd4b4dc7f14ee4320094d8b41684abff8539",
+  "base_composite_sha256": "b7dc3d5cef56c5ed8e03e9c54bde781b560897151e6b6fadc55ed40a22feaaa2",
+  "base_lock_sha256": "0144c9767921f51f37167784380029c01fc83caf5c5b54e803671bfc3c432bdf",
+  "tokenizer": "Use the tokenizer from the exact pinned base revision; no tokenizer files are bundled.",
+  "source": "TRAINING_CONFIG.json and training/COMMIT.json in the frozen release evidence"
+}

PRIMARY_SELECTION.json

@@ -0,0 +1,202 @@
+{
+  "schema": "mt11-high-heavy-l45-primary-selection/1",
+  "status": "FINAL_FROZEN",
+  "date": "2026-08-16",
+  "selection": {
+    "primary": "HIGH_HEAVY_L45",
+    "downgraded": "MEDCONF_MED34_HIGH30_CENTERED_S2",
+    "downgrade_evidence": {
+      "fresh_correct": 29,
+      "fresh_n": 240,
+      "vs_fresh_base_delta_correct": 9,
+      "question_cluster_bootstrap_95ci_delta_correct": [
+        -3,
+        22
+      ],
+      "significantly_above_fresh_base": false
+    },
+    "owner_override": {
+      "applied": true,
+      "prior_report_verdict": "DO_NOT_STOP_TRAINING_MIXED_OR_UNSTABLE",
+      "reason": "The prior report verdict combined HIGH and MEDCONF. The owner had predeclared that fresh-seed reproduction of the 34-point result ends training; HIGH exactly reproduced 34/240 and remained significantly above Fresh Base."
+    }
+  },
+  "high_complete_evidence": {
+    "no_cherry_pick": true,
+    "all_seeds_included": [
+      42,
+      43,
+      44,
+      45,
+      46,
+      47,
+      48,
+      49
+    ],
+    "pairing_note": "Different eval seeds are not sample-paired; comparisons use question aggregation. Combined health metrics are exhaustive sample totals.",
+    "original_block": {
+      "seeds": [
+        42,
+        43,
+        44,
+        45
+      ],
+      "n": 240,
+      "correct": 34,
+      "correct_by_seed": {
+        "42": 8,
+        "43": 9,
+        "44": 6,
+        "45": 11
+      },
+      "cap": {
+        "n": 91,
+        "pct": 37.917
+      },
+      "decoded_complete_box": {
+        "n": 150,
+        "pct": 62.5
+      },
+      "natural_stop": {
+        "n": 149,
+        "pct": 62.083
+      }
+    },
+    "fresh_block": {
+      "seeds": [
+        46,
+        47,
+        48,
+        49
+      ],
+      "n": 240,
+      "correct": 34,
+      "correct_by_seed": {
+        "46": 8,
+        "47": 9,
+        "48": 9,
+        "49": 8
+      },
+      "cap": {
+        "n": 84,
+        "pct": 35.0
+      },
+      "decoded_complete_box": {
+        "n": 151,
+        "pct": 62.917
+      },
+      "natural_stop": {
+        "n": 156,
+        "pct": 65.0
+      },
+      "vs_fresh_base": {
+        "fresh_base_correct": 20,
+        "fresh_base_n": 240,
+        "delta_correct": 14,
+        "question_cluster_bootstrap_95ci_delta_correct": [
+          2,
+          27
+        ],
+        "significantly_above": true
+      }
+    },
+    "combined_block": {
+      "seeds": [
+        42,
+        43,
+        44,
+        45,
+        46,
+        47,
+        48,
+        49
+      ],
+      "n": 480,
+      "correct": {
+        "n": 68,
+        "pct": 14.167
+      },
+      "cap": {
+        "n": 175,
+        "pct": 36.458
+      },
+      "decoded_complete_box": {
+        "n": 301,
+        "pct": 62.708
+      },
+      "natural_stop": {
+        "n": 305,
+        "pct": 63.542
+      }
+    }
+  },
+  "frozen_identity": {
+    "adapter_path": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/arms_90r6pp/HIGH_HEAVY_L45/train/checkpoint-update-00000024-tokens-000000409131/adapter",
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29",
+    "base_path": "/workspace/MATH-NOTHINK-AIME-P1-v1.0/model/mn8_eos_r1_2/Qwen3-4B-Base-906bfd4",
+    "tokenizer_path": "/workspace/MATH-NOTHINK-AIME-P1-v1.0/model/mn8_eos_r1_2/Qwen3-4B-Base-906bfd4",
+    "dynamic_lora": true,
+    "base_adapter_separate": true,
+    "merge": false,
+    "bake": false
+  },
+  "standard_think_evalscope_contract": {
+    "mode": "think",
+    "enable_thinking": true,
+    "framework": "EvalScope",
+    "score_authority": "EvalScope reviews",
+    "suites": [
+      "AIME24",
+      "AIME25"
+    ],
+    "independent_questions": 60,
+    "samples_per_four_seed_block": 240,
+    "sampling": {
+      "temperature": 0.6,
+      "top_p": 0.95,
+      "top_k": 20
+    },
+    "max_model_len": 32768,
+    "reserve_tokens": 64,
+    "max_new_tokens_rule": "32768 - exact_upstream_tokenize_prompt_count - 64",
+    "statistics": "question-clustered; do not treat 240 samples as 240 independent questions",
+    "serving": "vLLM dynamic LoRA mount; no merge or bake"
+  },
+  "route_freeze": {
+    "stop_all_new_mt11_training": true,
+    "short_cot": "FAIL",
+    "med34_score_uplift_r1_family_gate": "FAIL",
+    "automatic_continuation_forbidden": true,
+    "automatic_arm_expansion_forbidden": true,
+    "automatic_warm_start_forbidden": true,
+    "automatic_lora_combination_forbidden": true,
+    "historical_result_cards_modified": false,
+    "training_started_by_this_action": false,
+    "evaluation_started_by_this_action": false
+  },
+  "mn9_crossprotocol_eval": {
+    "approved_eval_only": true,
+    "must_not_stop_or_interfere": true,
+    "may_not_trigger_training_automatically": true,
+    "may_not_trigger_lora_combination_automatically": true,
+    "separate_approval_required_for_any_followup": true
+  },
+  "supervisor_read_only_check": {
+    "mt11_training_running": false,
+    "action_taken": "NONE",
+    "running_approved_eval_only_services": [
+      "ddc-mt11-mn9-crossprotocol-controller",
+      "ddc-mt11-mn9-crossprotocol-eval-disk-guard",
+      "ddc-mt11-mn9-crossprotocol-eval-gpu0",
+      "ddc-mt11-mn9-crossprotocol-eval-gpu1",
+      "ddc-mt11-mn9-crossprotocol-eval-gpu2",
+      "ddc-mt11-mn9-crossprotocol-eval-gpu3"
+    ]
+  },
+  "authoritative_sources": [
+    "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/MT11_TWO34_FRESHSEED_REVAL_FINAL.md",
+    "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/MT11_TWO34_FRESHSEED_REVAL_FINAL.json",
+    "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/RESULT_CARD_HIGH_HEAVY_L45.json",
+    "/workspace/DDC-v3.1/runs/MED34_SCORE_UPLIFT_CANARY_V1R1_SHARED_PREFIX/MED34_SCORE_UPLIFT_REPORT.md"
+  ]
+}

PRIMARY_SELECTION.md

@@ -0,0 +1,48 @@
+# MT11 HIGH_HEAVY_L45 主候选最终冻结登记
+
+**状态：FINAL_FROZEN · PRIMARY_SELECTED · STOP_ALL_NEW_MT11_TRAINING · 2026-08-16**
+
+## 最终决定
+
+- 当前 MT11 主候选：`HIGH_HEAVY_L45`。
+- `MEDCONF_MED34_HIGH30_CENTERED_S2` 降级：fresh seeds 46–49 为 29/240，且相对 Fresh Base 的题目级聚合 CI `[-3, 22]` 不排除 0。
+- 用户预先规定“两份 34 换全新 seed 仍高分即不再训练”；HIGH 在完整 fresh block 上精确复现 34/240，故停止所有新的 MT11 训练路线。
+- 权威 fresh-seed 总报告中的旧机器判决 `DO_NOT_STOP_TRAINING_MIXED_OR_UNSTABLE` 是对 HIGH 与 MEDCONF 的联合判决；本登记不修改该历史卡，而是按用户已明确的最终停止规则作 owner override。
+
+## HIGH 完整证据（不 cherry-pick）
+
+统计覆盖 AIME24+25 全部 60 题、每题四 seed；两组 seed 不做 sample 伪配对。
+
+| block | seeds | n | correct | cap | decoded complete box | natural stop |
+|---|---|---:|---:|---:|---:|---:|
+| original | 42–45 | 240 | 34（8/9/6/11） | 91（37.917%） | 150（62.500%） | 149（62.083%） |
+| fresh | 46–49 | 240 | 34（8/9/9/8） | 84（35.000%） | 151（62.917%） | 156（65.000%） |
+| combined | 42–49 | 480 | **68（14.167%）** | **175（36.458%）** | **301（62.708%）** | **305（63.542%）** |
+
+Fresh HIGH 相对 Fresh Base 20/240：`+14`，题目级聚合 95% CI `[2, 27]`，显著高于 Base。Original 与 fresh 均完整纳入 combined block；未筛 seed、题目或样本。
+
+## 冻结身份与合同
+
+- adapter：`/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/arms_90r6pp/HIGH_HEAVY_L45/train/checkpoint-update-00000024-tokens-000000409131/adapter`
+- adapter SHA256：`628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29`
+- base/tokenizer：`/workspace/MATH-NOTHINK-AIME-P1-v1.0/model/mn8_eos_r1_2/Qwen3-4B-Base-906bfd4`
+- serving：dynamic LoRA；base 与 adapter 分离；**不 merge、不 bake**。
+- 标准合同：Think、EvalScope reviews、AIME24+25、60 题、四 eval seeds、temperature 0.6、top_p 0.95、top_k 20、32K prompt-aware 上限 `32768 - exact_prompt_tokens - 64`；统计以题目为 cluster。
+
+## 路线冻结
+
+- 停止所有新的 MT11 训练路线；不得自动续训、扩臂、warm-start、组合 LoRA 或追加 training seed。
+- 短 CoT family gate 已 **FAIL**。
+- MED34 score-uplift R1 family gate 已 **FAIL**。
+- 不修改历史结果卡，不物化 merged/baked 模型，本次不启动训练或评测。
+- 已批准的 MN9 跨协议任务仅为纯评测，不得停止或干扰；其结果不得自动触发训练、扩臂或 LoRA 组合，任何后续须另行审批。
+
+## Supervisor 只读检查
+
+检查时无 MT11 训练 Supervisor 处于 `RUNNING`。正在运行的是已批准的 `ddc-mt11-mn9-crossprotocol-*` controller / eval / disk-guard 纯评测服务，未作停止或干扰。
+
+## 权威来源
+
+- `/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/MT11_TWO34_FRESHSEED_REVAL_FINAL.md`
+- `/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/MT11_TWO34_FRESHSEED_REVAL_FINAL.json`
+- `/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/RESULT_CARD_HIGH_HEAVY_L45.json`

R2_BACKUP_VERIFIED.json

@@ -0,0 +1,27 @@
+{
+  "schema": "mt11-r2-backup-verified/1",
+  "status": "PASS",
+  "candidate_id": "HIGH_HEAVY_L45",
+  "verified_at_utc": "2026-08-15T18:31:23Z",
+  "remote": "r2mtx:mxx/DDC-v3.1/releases/MT11-HIGH-HEAVY-L45-PRIMARY-v1",
+  "verification": {
+    "method": "rclone check --download --one-way",
+    "exit_code": 0,
+    "source_receipt": "REMOTE_VERIFICATION.json",
+    "remote_payload_file_count": 18,
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+  },
+  "local_asset": {
+    "path": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/arms_90r6pp/HIGH_HEAVY_L45/train/checkpoint-update-00000024-tokens-000000409131/adapter",
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29",
+    "sha_matches_authority": true,
+    "directory_mode": "0o555",
+    "weights_mode": "0o444"
+  },
+  "cleanup_protection": {
+    "status": "ACTIVE",
+    "source_receipt": "protection/PROTECTION_REGISTRATION.json"
+  },
+  "freeze_status": "ASSET_FROZEN_PROVENANCE_SCHEMA_MISSING",
+  "provenance_disclosure": "Training-time schema copy is missing; the current environment schema is not represented as the training-time schema."
+}

R2_FREEZE_STATUS.json

@@ -0,0 +1,41 @@
+{
+  "schema": "mt11-primary-candidate-freeze-status/1",
+  "status": "ASSET_FROZEN_PROVENANCE_SCHEMA_MISSING",
+  "candidate_id": "HIGH_HEAVY_L45",
+  "finalized_at_utc": "2026-08-15T18:29:03Z",
+  "asset_preservation": {
+    "complete": true,
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29",
+    "adapter_config_commit_verified": true,
+    "original_and_fresh_eval_cards_consistent": true,
+    "source_adapter_left_in_place": true,
+    "trainer_state_included": false
+  },
+  "provenance": {
+    "complete": false,
+    "expected_schema_sha256": "1819b4a0abf92ac070cf8bb84a47b4a7b8e72a537df2bee81687ae8f1a31a527",
+    "expected_schema_bytes": 15194,
+    "expected_schema_copy_found": false,
+    "current_environment_schema_sha256": "3cd276d4c4e2b00dfeb62e6b73802d6e403396bd4f23044b88ab8359791b36a2",
+    "current_environment_schema_bytes": 18084,
+    "current_copy_role": "environment_snapshot_only"
+  },
+  "remote": {
+    "path": "r2mtx:mxx/DDC-v3.1/releases/MT11-HIGH-HEAVY-L45-PRIMARY-v1",
+    "scope": "asset_and_evidence",
+    "verification": "PASS_RCLONE_CHECK_DOWNLOAD_ONE_WAY",
+    "matching_files": 20,
+    "differences": 0,
+    "verification_file": "R2_BACKUP_VERIFIED.json"
+  },
+  "readonly": {
+    "source_adapter_files": true,
+    "source_adapter_directory_traversable": true,
+    "freeze_package_files": true,
+    "freeze_package_directories_traversable": true,
+    "freeze_manifest": true,
+    "protection_markers": true
+  },
+  "cleanup_protection": "ACTIVE_REGISTERED_IN_DISK_HEADROOM_POLICY",
+  "claim_not_made": "FULLY_FROZEN_PROVENANCE_VERIFIED"
+}

README.md

@@ -0,0 +1,71 @@
+---
+base_model: Qwen/Qwen3-4B-Base
+library_name: peft
+tags:
+- peft
+- lora
+- qwen3
+- math
+- evalscope
+---
+# MT11-HIGH-L45 — v1.0.0
+
+## 中文说明
+
+这是 `HIGH_HEAVY_L45` 的**未合并动态 LoRA** 发布件：LoRA rank `r=64`、`alpha=128`，base 与 adapter 分离，未 merge、未 bake。请勿把本仓库当作完整基础模型。
+
+### 基础模型与固定身份
+
+- 基础模型：`Qwen/Qwen3-4B-Base`
+- 训练时精确本地路径：`/workspace/MATH-NOTHINK-AIME-P1-v1.0/model/mn8_eos_r1_2/Qwen3-4B-Base-906bfd4`
+- 固定 revision：`906bfd4b4dc7f14ee4320094d8b41684abff8539`
+- base composite SHA256：`b7dc3d5cef56c5ed8e03e9c54bde781b560897151e6b6fadc55ed40a22feaaa2`
+- base lock SHA256：`0144c9767921f51f37167784380029c01fc83caf5c5b54e803671bfc3c432bdf`
+- tokenizer：使用上述精确 base revision 自带 tokenizer；本仓库不重复上传 tokenizer 文件。
+- adapter SHA256：`628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29`
+
+### 评测合同与结果
+
+MT11 标准合同为 **Think + EvalScope reviews + dynamic LoRA + 32K**：AIME24+AIME25 共 60 题，temperature 0.6、top_p 0.95、top_k 20，输出上限为 `32768 - exact_prompt_tokens - 64`，统计以题目为 cluster，不把 240 个采样视为 240 道独立题。
+
+- seeds 42–45：`34/240`
+- fresh seeds 46–49：`34/240`
+- combined seeds 42–49：`68/480`（14.167%）
+- fresh health：cap 35.000%，decoded complete box 62.917%，natural stop 65.000%
+- combined health：cap 36.458%，decoded complete box 62.708%，natural stop 63.542%
+
+### 来源完整性披露
+
+状态为 **`ASSET_FROZEN_PROVENANCE_SCHEMA_MISSING`**。缺失的是训练当时 SHA256 为 `1819b4a0abf92ac070cf8bb84a47b4a7b8e72a537df2bee81687ae8f1a31a527` 的 schema 字节副本；现环境 schema 只能作为环境快照，不能冒充训练时副本。adapter 字节、adapter config/COMMIT 绑定及两组历史评测已独立核验，但**不保证精确重训复现**。
+
+### 动态 LoRA 使用示例
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+base_id = "Qwen/Qwen3-4B-Base"
+base_revision = "906bfd4b4dc7f14ee4320094d8b41684abff8539"
+adapter_id = "modrill/MT11-HIGH-L45"
+revision = "v1.0.0"
+
+tokenizer = AutoTokenizer.from_pretrained(base_id, revision=base_revision)
+base = AutoModelForCausalLM.from_pretrained(base_id, revision=base_revision, torch_dtype="auto", device_map="auto")
+model = PeftModel.from_pretrained(base, adapter_id, revision=revision)  # dynamic, unmerged
+```
+
+### 数据与许可证限制
+
+本发布不包含训练数据、原始 responses、trainer state 或 optimizer。现有冻结证据未授权为该 adapter 编造或新增独立许可证，因此本仓库**不声明新的 adapter license**。使用者必须自行核对并遵守基础模型、训练数据来源及适用数据/竞赛内容条款；评测结果不构成许可授权。
+
+## English
+
+This is the **unmerged dynamic LoRA** release of `HIGH_HEAVY_L45` (`r=64`, `alpha=128`). The base and adapter stay separate; nothing is merged or baked.
+
+The exact base is `Qwen/Qwen3-4B-Base` at revision `906bfd4b4dc7f14ee4320094d8b41684abff8539` (training path and hashes are listed above). Use that revision's tokenizer. The adapter SHA256 is `628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29`.
+
+Under the MT11 Think / EvalScope / dynamic 32K contract, seeds 42–45 scored 34/240, fresh seeds 46–49 scored 34/240, and the combined block scored 68/480. Fresh health: 35.000% cap, 62.917% decoded-complete-box, 65.000% natural stop. Combined health: 36.458%, 62.708%, and 63.542%, respectively.
+
+Provenance status is **`ASSET_FROZEN_PROVENANCE_SCHEMA_MISSING`**: the training-time schema byte copy is missing. The available current-environment schema is not represented as the historical training schema. Adapter bytes and evaluation evidence were verified, but exact retraining reproducibility is **not guaranteed**.
+
+No training data, raw responses, trainer state, optimizer state, or secrets are included. Existing project evidence does not establish a new standalone license for this adapter, so none is invented here. Users must review and comply with the base-model and applicable data/source terms.

RESULT_CARD_FRESH_SEEDS_46_49.json

@@ -0,0 +1,171 @@
+{
+  "schema": "mt11-two34-freshseed-reval-result-card/1",
+  "status": "DONE",
+  "arm_id": "HIGH_HEAVY_L45",
+  "eval_seeds": [
+    46,
+    47,
+    48,
+    49
+  ],
+  "training": {
+    "dynamic_lora": true,
+    "merge_or_bake": false,
+    "adapter": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/arms_90r6pp/HIGH_HEAVY_L45/train/checkpoint-update-00000024-tokens-000000409131/adapter",
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29",
+    "reused_frozen_adapter": true,
+    "this_round_trained": false
+  },
+  "evaluation": {
+    "contract": "Think EvalScope dynamic 32K, AIME24+25, seeds 46-49",
+    "score_authority": "EvalScope reviews",
+    "correct": 34,
+    "n": 240,
+    "accuracy": 0.14166666666666666,
+    "by_seed": {
+      "46": {
+        "correct": 8,
+        "n": 60
+      },
+      "47": {
+        "correct": 9,
+        "n": 60
+      },
+      "48": {
+        "correct": 9,
+        "n": 60
+      },
+      "49": {
+        "correct": 8,
+        "n": 60
+      }
+    },
+    "by_year": {
+      "aime24": {
+        "correct": 18,
+        "n": 120
+      },
+      "aime25": {
+        "correct": 16,
+        "n": 120
+      }
+    },
+    "majority_at_4": 6,
+    "pass_at_4": 14,
+    "stable_correct_4_of_4": 3,
+    "health": {
+      "n": 240,
+      "cap_n": 84,
+      "cap_pct": 35.0,
+      "decoded_complete_box_n": 151,
+      "decoded_complete_box_pct": 62.917,
+      "natural_stop_n": 156,
+      "natural_stop_pct": 65.0,
+      "native_eos_151643_n": 156,
+      "protocol_eos_151645_n": 0,
+      "misaligned_stop_n": 0,
+      "mean_completion_tokens": 13002.058333333332,
+      "completion_tokens_p50": 2008,
+      "completion_tokens_p95": 32609
+    }
+  },
+  "original_42_45": {
+    "correct": 34,
+    "by_seed": {
+      "42": 8,
+      "43": 9,
+      "44": 6,
+      "45": 11
+    },
+    "health": {
+      "cap_pct": 37.917,
+      "decoded_complete_box_pct": 62.5,
+      "natural_stop_pct": 62.083,
+      "majority_at_4": 8,
+      "pass_at_4": 16
+    },
+    "majority_at_4": 8
+  },
+  "question_aggregated_vs_original_42_45": {
+    "pairing": "question_aggregated_k_of_4_not_same_seed",
+    "observed_delta_correct_240": 0,
+    "mean_question_rate_delta": 0.0,
+    "question_cluster_bootstrap_95ci_delta_correct_240": [
+      -12.0,
+      12.0
+    ],
+    "bootstrap_positive_fraction": 0.47215,
+    "ci_excludes_zero_positive": false,
+    "ci_excludes_zero_negative": false,
+    "majority": {
+      "new_majority_at_4": 6,
+      "old_majority_at_4": 8,
+      "delta": -2,
+      "questions_gained_majority": 2,
+      "questions_lost_majority": 4,
+      "mcnemar_note": "question-level majority flip; seeds differ so this is not sample pairing",
+      "mcnemar_exact_two_sided_p": 0.6875
+    }
+  },
+  "question_aggregated_vs_fresh_base_20": {
+    "pairing": "question_aggregated_k_of_4_not_same_seed",
+    "observed_delta_correct_240": 14,
+    "mean_question_rate_delta": 0.058333333333333334,
+    "question_cluster_bootstrap_95ci_delta_correct_240": [
+      2.0,
+      27.0
+    ],
+    "bootstrap_positive_fraction": 0.98715,
+    "ci_excludes_zero_positive": true,
+    "ci_excludes_zero_negative": false,
+    "majority": {
+      "new_majority_at_4": 6,
+      "old_majority_at_4": 0,
+      "delta": 6,
+      "questions_gained_majority": 6,
+      "questions_lost_majority": 0,
+      "mcnemar_note": "question-level majority flip; seeds differ so this is not sample pairing",
+      "mcnemar_exact_two_sided_p": 0.03125
+    },
+    "fresh_base_correct_240": 20,
+    "note": "Fresh Base seeds are 42-45; comparison is question-level only"
+  },
+  "decision": {
+    "still_above_fresh_base_20": true,
+    "still_near_34": true,
+    "collapsed_near_base": false,
+    "new_correct_240": 34,
+    "gates": {
+      "new_correct_240_min_for_high": 28,
+      "fresh_base_ci_lower_bound_must_gt_0": true
+    }
+  },
+  "artifacts": {
+    "shards": [
+      {
+        "lane_id": "HIGH_HEAVY_L45_S4647",
+        "seeds": [
+          46,
+          47
+        ],
+        "summary": "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/eval/HIGH_HEAVY_L45_S4647/HIGH_HEAVY_L45/think/SUMMARY.json",
+        "eval_contract": "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/eval/HIGH_HEAVY_L45_S4647/HIGH_HEAVY_L45/think/EVAL_CONTRACT.json",
+        "correct": 17,
+        "n": 120,
+        "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+      },
+      {
+        "lane_id": "HIGH_HEAVY_L45_S4849",
+        "seeds": [
+          48,
+          49
+        ],
+        "summary": "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/eval/HIGH_HEAVY_L45_S4849/HIGH_HEAVY_L45/think/SUMMARY.json",
+        "eval_contract": "/workspace/DDC-v3.1/runs/mt11_two34_freshseed_reval_20260815/eval/HIGH_HEAVY_L45_S4849/HIGH_HEAVY_L45/think/EVAL_CONTRACT.json",
+        "correct": 17,
+        "n": 120,
+        "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+      }
+    ]
+  }
+}

RESULT_CARD_ORIGINAL_SEEDS_42_45.json

@@ -0,0 +1,92 @@
+{
+  "schema": "mt11-r2-first-wave-result-card/1",
+  "status": "DONE",
+  "arm_id": "HIGH_HEAVY_L45",
+  "training": {
+    "completed_optimizer_updates": 24,
+    "unique_rows": 90,
+    "base_parent_bitwise_unchanged": true,
+    "dynamic_lora": true,
+    "merge_or_bake": false,
+    "adapter": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/arms_90r6pp/HIGH_HEAVY_L45/train/checkpoint-update-00000024-tokens-000000409131/adapter",
+    "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+  },
+  "evaluation": {
+    "score_authority": "EvalScope reviews",
+    "correct": 34,
+    "n": 240,
+    "accuracy": 0.14166666666666666,
+    "by_seed": {
+      "42": {
+        "correct": 8,
+        "n": 60
+      },
+      "43": {
+        "correct": 9,
+        "n": 60
+      },
+      "44": {
+        "correct": 6,
+        "n": 60
+      },
+      "45": {
+        "correct": 11,
+        "n": 60
+      }
+    },
+    "by_year": {
+      "aime24": {
+        "correct": 15,
+        "n": 120
+      },
+      "aime25": {
+        "correct": 19,
+        "n": 120
+      }
+    },
+    "majority_at_4": 8,
+    "pass_at_4": 16,
+    "stable_correct_4_of_4": 1,
+    "question_cluster_bootstrap_95ci_correct_240": [
+      19.0,
+      51.0
+    ],
+    "health": {
+      "n": 240,
+      "cap_n": 91,
+      "cap_pct": 37.917,
+      "decoded_complete_box_n": 150,
+      "decoded_complete_box_pct": 62.5,
+      "natural_stop_n": 149,
+      "natural_stop_pct": 62.083,
+      "native_eos_151643_n": 149,
+      "protocol_eos_151645_n": 0,
+      "misaligned_stop_n": 0,
+      "mean_completion_tokens": 13464.795833333334,
+      "completion_tokens_p50": 1930,
+      "completion_tokens_p95": 32604
+    }
+  },
+  "references": {
+    "fresh_base_contract_aligned": {
+      "correct": 20,
+      "n": 240,
+      "delta_correct": 14,
+      "item_level_pairing_available": false,
+      "reason": "The retained fresh-base artifact has SUMMARY/captures but no EvalScope review rows."
+    },
+    "historical_R2_frozen_reference": {
+      "correct": 30,
+      "n": 240,
+      "delta_correct": 4,
+      "contract_source": "/workspace/DDC-v3.1/runs/mt11_r2_refinement_20260814/SCORE_FIRST_GATES.json"
+    }
+  },
+  "artifacts": {
+    "summary": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/eval/first_wave/HIGH_HEAVY_L45/HIGH_HEAVY_L45/think/SUMMARY.json",
+    "eval_contract": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/eval/first_wave/HIGH_HEAVY_L45/HIGH_HEAVY_L45/think/EVAL_CONTRACT.json",
+    "prompt_preflight": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/eval/first_wave/HIGH_HEAVY_L45/HIGH_HEAVY_L45/think/PROMPT_PREFLIGHT.json",
+    "raw_captures": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/eval/first_wave/HIGH_HEAVY_L45/HIGH_HEAVY_L45/think/captures",
+    "evalscope_results": "/workspace/DDC-v3.1/runs/mt11_r2_384k96_execution_20260814/eval/first_wave/HIGH_HEAVY_L45/HIGH_HEAVY_L45/think/results"
+  }
+}

SHA256SUMS

@@ -0,0 +1,11 @@
+24187418cc1f772de8cfda265327683f7f8c5ab6c3e346d70b77598217b507d2  BASE_MODEL_REFERENCE.json
+f7f4861cf82f0ba705a0f3e5a7a20e2427b37e6adfb8ba48f29669eaf49a386e  PRIMARY_SELECTION.json
+a0d1e2adfb94feef0ce01fccffbd42ba7f48a6c0e78e7bcb08145b5adf056535  PRIMARY_SELECTION.md
+3fb9068947abba5bd4d4e6f7746576438e45376920c7167cd311215ae828575a  R2_BACKUP_VERIFIED.json
+52f1d6589071366b2da9efdda5339229275b06f4f58b6b386134c44fd6e277d4  R2_FREEZE_STATUS.json
+3db5ee30b32e294cb7ce0474d7f0cca6d0bf21af39e20412eb762066aaaeae13  README.md
+6b54852818f1ec8d28a491b521283db5598da95bd9e7780247f35b3748382a35  RESULT_CARD_FRESH_SEEDS_46_49.json
+9c070861760d786a0ccde710eead98257d1fd14339a82ddf35e8908d0fcb1dfe  RESULT_CARD_ORIGINAL_SEEDS_42_45.json
+ba62c8fdc1cd19e725cf3faeac9f9075aade701c2b05d23bd010eec0e2314eb7  SHA256_MANIFEST.json
+0a382e55427ff1bca424645ea25c0498106f1d622b34115b1b95f1255e629f2c  adapter_config.json
+628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29  adapter_model.safetensors

SHA256_MANIFEST.json

@@ -0,0 +1,59 @@
+{
+  "schema": "mt11-hf-stage-sha256-manifest/1",
+  "repo_id": "modrill/MT11-HIGH-L45",
+  "version": "v1.0.0",
+  "created_at_utc": "2026-08-15T18:31:23Z",
+  "files": [
+    {
+      "path": "BASE_MODEL_REFERENCE.json",
+      "bytes": 628,
+      "sha256": "24187418cc1f772de8cfda265327683f7f8c5ab6c3e346d70b77598217b507d2"
+    },
+    {
+      "path": "PRIMARY_SELECTION.json",
+      "bytes": 5686,
+      "sha256": "f7f4861cf82f0ba705a0f3e5a7a20e2427b37e6adfb8ba48f29669eaf49a386e"
+    },
+    {
+      "path": "PRIMARY_SELECTION.md",
+      "bytes": 3222,
+      "sha256": "a0d1e2adfb94feef0ce01fccffbd42ba7f48a6c0e78e7bcb08145b5adf056535"
+    },
+    {
+      "path": "R2_BACKUP_VERIFIED.json",
+      "bytes": 1188,
+      "sha256": "3fb9068947abba5bd4d4e6f7746576438e45376920c7167cd311215ae828575a"
+    },
+    {
+      "path": "R2_FREEZE_STATUS.json",
+      "bytes": 1603,
+      "sha256": "52f1d6589071366b2da9efdda5339229275b06f4f58b6b386134c44fd6e277d4"
+    },
+    {
+      "path": "README.md",
+      "bytes": 4244,
+      "sha256": "3db5ee30b32e294cb7ce0474d7f0cca6d0bf21af39e20412eb762066aaaeae13"
+    },
+    {
+      "path": "RESULT_CARD_FRESH_SEEDS_46_49.json",
+      "bytes": 4991,
+      "sha256": "6b54852818f1ec8d28a491b521283db5598da95bd9e7780247f35b3748382a35"
+    },
+    {
+      "path": "RESULT_CARD_ORIGINAL_SEEDS_42_45.json",
+      "bytes": 2960,
+      "sha256": "9c070861760d786a0ccde710eead98257d1fd14339a82ddf35e8908d0fcb1dfe"
+    },
+    {
+      "path": "adapter_config.json",
+      "bytes": 1088,
+      "sha256": "0a382e55427ff1bca424645ea25c0498106f1d622b34115b1b95f1255e629f2c"
+    },
+    {
+      "path": "adapter_model.safetensors",
+      "bytes": 528550256,
+      "sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+    }
+  ],
+  "adapter_sha256": "628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29"
+}

adapter_config.json

@@ -0,0 +1,46 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": null,
+  "base_model_name_or_path": "Qwen/Qwen3-4B-Base",
+  "bias": "none",
+  "corda_config": null,
+  "ensure_weight_tying": false,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 128,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 64,
+  "rank_pattern": {},
+  "revision": "906bfd4b4dc7f14ee4320094d8b41684abff8539",
+  "target_modules": [
+    "k_proj",
+    "q_proj",
+    "up_proj",
+    "down_proj",
+    "gate_proj",
+    "o_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:628a5b4c40584c524479b80773166f49b5c49d7223b34d4cf254064d6b65bd29
+size 528550256