{
  "schema_version": 1,
  "project": "rl",
  "title": "把最终补丁反馈，分配到编辑轮次",
  "recorded_at": "2026-11-05T02:35:19+08:00",
  "current": {
    "title": "把最终补丁反馈，分配到编辑轮次",
    "goal": "检验局部反事实反馈的分配是否改善代理学习",
    "direction": "Counterfactual Edit Credit for Long-Horizon Software Agents",
    "status": "方法稿 · 效果待测",
    "reliable": "320 条开发轨迹中，244 次删除查询可用，92 次有非零差异。[[R1]]",
    "summary": "11-05重新核对：旧O42-retry1账本仍停在34/80次完成调用；step30 HF结构存在且通过检查，但未有新恢复release。两机GPU有活动、父容器运行，不等于这条RL已恢复。方法和开发pilot的既有证据保留，主要M−B策略效果仍未测量。[[RESUME_1105]][[R1]][[R4]]",
    "next": "先确认当前资源归属与调度责任，关闭外部终止后的恢复和评测准入缺口，再按冻结协议推进比较。[[RESUME_1105]][[RECOVERY_PROTOCOL]]",
    "observed_at": "2026-11-05T02:25:51+08:00",
    "recorded_at": "2026-11-05T02:31:44+08:00",
    "recent_changes": [
      "11-05：消费账本34完成/35准备；step30四分片427张量结构通过，恢复仍未放行。[[RESUME_1105]]",
      "11-05：两机资源活动未归属到旧O，未启停或清理任何任务。[[RESUME_1105]]",
      "10-01：旧B首步判分污染和raw batch证据缺口保持，未用旧reward填主结果。[[B_PREFLIGHT]][[RAW_GAP]]"
    ],
    "constraints": [
      "原始结果、失败和unknown保留，不将缺判分当0。",
      "开始计算任务前核对当前进程、资源与准入收据。",
      "来源变动后核对受影响主张，网页不替代原始证据。",
      "主要终点80、固定种子/任务/评测协议，不按在线reward选择终点。",
      "gpu23外部任务不得停止；资源是否空闲必须现场确认。"
    ],
    "conflicts": [
      {
        "id": "RL-C001",
        "title": "旧交接正文仍写“19/80活跃”，与后续终止和恢复门冲突",
        "historical": "09-28早期19/80运行描述",
        "current": "09-28完成34次后终止；10-01材料未放行新恢复。",
        "resolution": "按时间与运行收据解释；旧文字保留为历史，不作为10-04现场状态。",
        "sources": [
          "R4",
          "RECOVERY_ARTIFACT",
          "HANDOFF"
        ]
      }
    ],
    "must_verify": [
      "当前可分配GPU、使用窗口与唯一训练调度负责人；11-05快照不是未来占用授权。",
      "新恢复release须重新绑定checkpoint、终止证据和两机preflight。",
      "正式评测准入和unknown规则仍需闭环；不读取封存标签来补齐接手状态。"
    ]
  },
  "documents": [
    {
      "id": "R1",
      "path": "paper-rl/paper/sections/5_evaluation.tex",
      "title": "现稿评测：pilot、成本与效果Pending",
      "kind": "research",
      "observed_at": "当前稿件",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "90a6a8574b27466b6e876253200ff9e7762f5747fc8c62cfe4cb0e71484bc811",
      "priority": 2,
      "public": true,
      "sha256": "90a6a8574b27466b6e876253200ff9e7762f5747fc8c62cfe4cb0e71484bc811",
      "reviewed_sha256": "90a6a8574b27466b6e876253200ff9e7762f5747fc8c62cfe4cb0e71484bc811",
      "bytes": 4078,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R1/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R1/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R2",
      "path": "paper-rl/PAPER_PLAN.md",
      "title": "方法与O/B/M/E实验计划",
      "kind": "research",
      "observed_at": "2026-09-27",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "f345575600d0c77bdd22b735a141a30a97a0513e3d65dbe0061cc37063e8cbe1",
      "priority": 2,
      "public": true,
      "sha256": "f345575600d0c77bdd22b735a141a30a97a0513e3d65dbe0061cc37063e8cbe1",
      "reviewed_sha256": "f345575600d0c77bdd22b735a141a30a97a0513e3d65dbe0061cc37063e8cbe1",
      "bytes": 7706,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R2/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R2/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R3",
      "path": "paper-rl/paper/sections/6_limitations.tex",
      "title": "现稿干预、优化与判分限制",
      "kind": "research",
      "observed_at": "当前稿件",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "c0d214dcd0f06f9efd49d21e934252f7c53c7e32cdfd1c949ca2e73967d72adc",
      "priority": 2,
      "public": true,
      "sha256": "c0d214dcd0f06f9efd49d21e934252f7c53c7e32cdfd1c949ca2e73967d72adc",
      "reviewed_sha256": "c0d214dcd0f06f9efd49d21e934252f7c53c7e32cdfd1c949ca2e73967d72adc",
      "bytes": 2252,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R3/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R3/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R4",
      "path": "paper-rl/results/O42_retry1_external_termination_20260928.json",
      "title": "O42-retry1外部终止收据",
      "kind": "result",
      "observed_at": "2026-09-28 15:35",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "5b9ce35674fd8114d171fb00d8b0336ffdd278b5f6e6bc9161e12d06244577f9",
      "priority": 2,
      "public": true,
      "sha256": "5b9ce35674fd8114d171fb00d8b0336ffdd278b5f6e6bc9161e12d06244577f9",
      "reviewed_sha256": "5b9ce35674fd8114d171fb00d8b0336ffdd278b5f6e6bc9161e12d06244577f9",
      "bytes": 1575,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R4/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R4/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R5",
      "path": "paper-rl/results/common_profile_runtime_coverage_final_20260930.json",
      "title": "500共同环境最终汇总（纠正后）",
      "kind": "result",
      "observed_at": "2026-10-01 纠正",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "45ef82e6cc1980428d679e28ad6f29ab4ec72fa843521448f25a5172a2385a48",
      "priority": 2,
      "public": true,
      "sha256": "45ef82e6cc1980428d679e28ad6f29ab4ec72fa843521448f25a5172a2385a48",
      "reviewed_sha256": "45ef82e6cc1980428d679e28ad6f29ab4ec72fa843521448f25a5172a2385a48",
      "bytes": 31698,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R5/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R5/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R6",
      "path": "paper-rl/PAPER_ACCEPTANCE_CONTRACT.md",
      "title": "草稿验收约定与结果版本缺口",
      "kind": "research",
      "observed_at": "2026-09-27",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "3d9ae3c2667e2c4a0bc56dbf24e83b279b8267c122956181e230d585242208cd",
      "priority": 2,
      "public": true,
      "sha256": "3d9ae3c2667e2c4a0bc56dbf24e83b279b8267c122956181e230d585242208cd",
      "reviewed_sha256": "3d9ae3c2667e2c4a0bc56dbf24e83b279b8267c122956181e230d585242208cd",
      "bytes": 3399,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R6/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R6/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R7",
      "path": "paper-rl/STATUS_20260928.md",
      "title": "当前交接：汇总更正与运行边界",
      "kind": "research",
      "observed_at": "2026-10-01 更新",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "824026dba6207f23184e0c9dd09e37f9e8ea21b534a21ce4b6a878a03db40b2f",
      "priority": 2,
      "public": true,
      "sha256": "824026dba6207f23184e0c9dd09e37f9e8ea21b534a21ce4b6a878a03db40b2f",
      "reviewed_sha256": "824026dba6207f23184e0c9dd09e37f9e8ea21b534a21ce4b6a878a03db40b2f",
      "bytes": 38313,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R7/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R7/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "R8",
      "path": "paper-rl/results/grading_recovery_paper_update_20260928.json",
      "title": "恢复限制稿局部审查与编译回执",
      "kind": "result",
      "observed_at": "2026-09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "7756cd4c431ab0056467e333e00fcdd6feafe4e39f6a71e6b0fbe4c31fc05b8e",
      "priority": 2,
      "public": true,
      "sha256": "7756cd4c431ab0056467e333e00fcdd6feafe4e39f6a71e6b0fbe4c31fc05b8e",
      "reviewed_sha256": "7756cd4c431ab0056467e333e00fcdd6feafe4e39f6a71e6b0fbe4c31fc05b8e",
      "bytes": 3077,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R8/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/R8/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "HANDOFF",
      "path": "rl/研究情况/20260927交接.md",
      "title": "跨日交接（含混合日期）",
      "kind": "handoff",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "49ed75c1115062c8bd9d8800629352ae582d6b344a8c98feff51ff5974140bd2",
      "priority": 1,
      "public": true,
      "sha256": "49ed75c1115062c8bd9d8800629352ae582d6b344a8c98feff51ff5974140bd2",
      "reviewed_sha256": "49ed75c1115062c8bd9d8800629352ae582d6b344a8c98feff51ff5974140bd2",
      "bytes": 95089,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/HANDOFF/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/HANDOFF/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "EVAL_PROTOCOL",
      "path": "paper-rl/EVALUATION_PROTOCOL_20260927.md",
      "title": "固定评测协议",
      "kind": "definition",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "9065159e575781bb3880cf3e2ad3368562b5a071c6ff1fd0765f642b76797a6b",
      "priority": 1,
      "public": true,
      "sha256": "9065159e575781bb3880cf3e2ad3368562b5a071c6ff1fd0765f642b76797a6b",
      "reviewed_sha256": "9065159e575781bb3880cf3e2ad3368562b5a071c6ff1fd0765f642b76797a6b",
      "bytes": 6088,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/EVAL_PROTOCOL/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/EVAL_PROTOCOL/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "RECOVERY_PROTOCOL",
      "path": "paper-rl/TRAINING_RECOVERY_PROTOCOL_20260927.md",
      "title": "训练恢复协议",
      "kind": "definition",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "307e44171b9ed260e3e24682ae0e5025fb58e4fa9714508572af99d9ead0e853",
      "priority": 1,
      "public": true,
      "sha256": "307e44171b9ed260e3e24682ae0e5025fb58e4fa9714508572af99d9ead0e853",
      "reviewed_sha256": "307e44171b9ed260e3e24682ae0e5025fb58e4fa9714508572af99d9ead0e853",
      "bytes": 3321,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RECOVERY_PROTOCOL/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RECOVERY_PROTOCOL/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "CONFIG_AUDIT",
      "path": "paper-rl/results/credit_arm_config_comparison_20261001.json",
      "title": "O/B/M入口配置核对",
      "kind": "receipt",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "1cbcc9d3fc923df43ddc41132dc4cc99ae070a0b349181fddeb97e4200b5646a",
      "priority": 1,
      "public": true,
      "sha256": "1cbcc9d3fc923df43ddc41132dc4cc99ae070a0b349181fddeb97e4200b5646a",
      "reviewed_sha256": "1cbcc9d3fc923df43ddc41132dc4cc99ae070a0b349181fddeb97e4200b5646a",
      "bytes": 6146,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CONFIG_AUDIT/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CONFIG_AUDIT/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "ENTROPY_AUDIT",
      "path": "paper-rl/results/credit_entropy_path_audit_20261001.json",
      "title": "B低熵与response mask核对",
      "kind": "receipt",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "ed387438f8803036d1eb5e9a5ddabaca425e27519d7b4f9246ed5a6dd10402a3",
      "priority": 1,
      "public": true,
      "sha256": "ed387438f8803036d1eb5e9a5ddabaca425e27519d7b4f9246ed5a6dd10402a3",
      "reviewed_sha256": "ed387438f8803036d1eb5e9a5ddabaca425e27519d7b4f9246ed5a6dd10402a3",
      "bytes": 1579,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/ENTROPY_AUDIT/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/ENTROPY_AUDIT/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "B_PREFLIGHT",
      "path": "paper-rl/results/credit_first_update_preflight_log_audit_20261001.json",
      "title": "B首更新前判分缺失审计",
      "kind": "receipt",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "070db41162466e39344762bdffd451c5e1c373978d1220f4574546614c39bf60",
      "priority": 1,
      "public": true,
      "sha256": "070db41162466e39344762bdffd451c5e1c373978d1220f4574546614c39bf60",
      "reviewed_sha256": "070db41162466e39344762bdffd451c5e1c373978d1220f4574546614c39bf60",
      "bytes": 2619,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/B_PREFLIGHT/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/B_PREFLIGHT/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "RECOVERY_ARTIFACT",
      "path": "paper-rl/results/O42_step30_recovery_artifact_integrity_20261001.json",
      "title": "终止后step30恢复材料",
      "kind": "receipt",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "969c877b41865f4490bd969c3594c7d5902781a3f2267517309c230a0fb2764b",
      "priority": 1,
      "public": true,
      "sha256": "969c877b41865f4490bd969c3594c7d5902781a3f2267517309c230a0fb2764b",
      "reviewed_sha256": "969c877b41865f4490bd969c3594c7d5902781a3f2267517309c230a0fb2764b",
      "bytes": 1563,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RECOVERY_ARTIFACT/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RECOVERY_ARTIFACT/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "B_PLAN",
      "path": "paper-rl/results/B_clean_rerun_preflight_plan_20261001.json",
      "title": "B受控重跑预检计划（未放行）",
      "kind": "definition",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "b3993436a5fd0f50eebd22bb36703c0c6879803f472ebc25f6f9370c1812f2fa",
      "priority": 1,
      "public": true,
      "sha256": "b3993436a5fd0f50eebd22bb36703c0c6879803f472ebc25f6f9370c1812f2fa",
      "reviewed_sha256": "b3993436a5fd0f50eebd22bb36703c0c6879803f472ebc25f6f9370c1812f2fa",
      "bytes": 1529,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/B_PLAN/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/B_PLAN/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "RAW_GAP",
      "path": "paper-rl/results/credit_raw_batch_evidence_gap_20261001.json",
      "title": "首步raw batch证据缺口",
      "kind": "receipt",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "c058f891c8bceffd67d2f11a4caaee5122cd7ff287693675627165252234a79b",
      "priority": 1,
      "public": true,
      "sha256": "c058f891c8bceffd67d2f11a4caaee5122cd7ff287693675627165252234a79b",
      "reviewed_sha256": "c058f891c8bceffd67d2f11a4caaee5122cd7ff287693675627165252234a79b",
      "bytes": 2498,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RAW_GAP/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RAW_GAP/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "CREDIT_CODE",
      "path": "rl/skyrl/swe_skyrl_recipe/edit_credit.py",
      "title": "编辑块选择与产物恢复",
      "kind": "code",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "6b019248c65f52c5675d2ebb8f629fcd47321494650ab13b5680628f58c12921",
      "priority": 2,
      "public": true,
      "sha256": "6b019248c65f52c5675d2ebb8f629fcd47321494650ab13b5680628f58c12921",
      "reviewed_sha256": "6b019248c65f52c5675d2ebb8f629fcd47321494650ab13b5680628f58c12921",
      "bytes": 16085,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CREDIT_CODE/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CREDIT_CODE/source.py",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TRAINER_CODE",
      "path": "rl/skyrl/swe_skyrl_recipe/credit_trainer.py",
      "title": "辅助信用接入训练",
      "kind": "code",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "c8fb5e4010719b951894df838b7b985ccd25f0df0f1720ee9ad199f2358b3d37",
      "priority": 2,
      "public": true,
      "sha256": "c8fb5e4010719b951894df838b7b985ccd25f0df0f1720ee9ad199f2358b3d37",
      "reviewed_sha256": "c8fb5e4010719b951894df838b7b985ccd25f0df0f1720ee9ad199f2358b3d37",
      "bytes": 8222,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TRAINER_CODE/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TRAINER_CODE/source.py",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "DIARY_0907",
      "path": "rl/研究情况/0907日报-接手后.md",
      "title": "09-07原始日报",
      "kind": "daily",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "af262e098390f449f5c9a66255a5bfb34b357dfcbc562a3c0898b0a281d4b912",
      "priority": 2,
      "public": true,
      "sha256": "af262e098390f449f5c9a66255a5bfb34b357dfcbc562a3c0898b0a281d4b912",
      "reviewed_sha256": "af262e098390f449f5c9a66255a5bfb34b357dfcbc562a3c0898b0a281d4b912",
      "bytes": 3649,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/DIARY_0907/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/DIARY_0907/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "DIARY_0908",
      "path": "rl/研究情况/0908日报-接手后.md",
      "title": "09-08原始日报",
      "kind": "daily",
      "observed_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "09b796844f169f11053c902e1e6d4b19eb282eb3c7deac8122188676f12e6598",
      "priority": 2,
      "public": true,
      "sha256": "09b796844f169f11053c902e1e6d4b19eb282eb3c7deac8122188676f12e6598",
      "reviewed_sha256": "09b796844f169f11053c902e1e6d4b19eb282eb3c7deac8122188676f12e6598",
      "bytes": 6871,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/DIARY_0908/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/DIARY_0908/source.md",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_0_ABSTRACT",
      "path": "paper-rl/paper/sections/0_abstract.tex",
      "title": "0_abstract · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "81446d374826f6419e0ad1449ccb1e5554882bea6ffc708fc45bace464f30d14",
      "priority": 2,
      "public": true,
      "sha256": "81446d374826f6419e0ad1449ccb1e5554882bea6ffc708fc45bace464f30d14",
      "reviewed_sha256": "81446d374826f6419e0ad1449ccb1e5554882bea6ffc708fc45bace464f30d14",
      "bytes": 1343,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_0_ABSTRACT/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_0_ABSTRACT/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_1_INTRODUCTION",
      "path": "paper-rl/paper/sections/1_introduction.tex",
      "title": "1_introduction · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "48f1d8352d3d4052860946a14bc7df25822183f7f661bb58bd74f074006954ad",
      "priority": 2,
      "public": true,
      "sha256": "48f1d8352d3d4052860946a14bc7df25822183f7f661bb58bd74f074006954ad",
      "reviewed_sha256": "48f1d8352d3d4052860946a14bc7df25822183f7f661bb58bd74f074006954ad",
      "bytes": 3348,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_1_INTRODUCTION/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_1_INTRODUCTION/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_2_RELATED_WORK",
      "path": "paper-rl/paper/sections/2_related_work.tex",
      "title": "2_related_work · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "77ab730fb26d9486811e44f5ebe1596843e6b2780d033580cc69e65cc1fab363",
      "priority": 2,
      "public": true,
      "sha256": "77ab730fb26d9486811e44f5ebe1596843e6b2780d033580cc69e65cc1fab363",
      "reviewed_sha256": "77ab730fb26d9486811e44f5ebe1596843e6b2780d033580cc69e65cc1fab363",
      "bytes": 5963,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_2_RELATED_WORK/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_2_RELATED_WORK/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_3_METHOD",
      "path": "paper-rl/paper/sections/3_method.tex",
      "title": "3_method · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "22957a8c2c5135eb9586d02e8885fa841c05768fa5d183fa3b97be338a2b2fa1",
      "priority": 2,
      "public": true,
      "sha256": "22957a8c2c5135eb9586d02e8885fa841c05768fa5d183fa3b97be338a2b2fa1",
      "reviewed_sha256": "22957a8c2c5135eb9586d02e8885fa841c05768fa5d183fa3b97be338a2b2fa1",
      "bytes": 7387,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_3_METHOD/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_3_METHOD/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_4_EXPERIMENTS",
      "path": "paper-rl/paper/sections/4_experiments.tex",
      "title": "4_experiments · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "36dc0821c114189c83c154f1686b7c08a289abbd1e7f5095443b91326d0c2e74",
      "priority": 2,
      "public": true,
      "sha256": "36dc0821c114189c83c154f1686b7c08a289abbd1e7f5095443b91326d0c2e74",
      "reviewed_sha256": "36dc0821c114189c83c154f1686b7c08a289abbd1e7f5095443b91326d0c2e74",
      "bytes": 4963,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_4_EXPERIMENTS/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_4_EXPERIMENTS/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_7_CONCLUSION",
      "path": "paper-rl/paper/sections/7_conclusion.tex",
      "title": "7_conclusion · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "90236c20bfcdd8242c44721b14ea020b117ef4e9cb0829bba3777395e367203d",
      "priority": 2,
      "public": true,
      "sha256": "90236c20bfcdd8242c44721b14ea020b117ef4e9cb0829bba3777395e367203d",
      "reviewed_sha256": "90236c20bfcdd8242c44721b14ea020b117ef4e9cb0829bba3777395e367203d",
      "bytes": 522,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_7_CONCLUSION/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_7_CONCLUSION/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "TEX_A_APPENDIX",
      "path": "paper-rl/paper/sections/A_appendix.tex",
      "title": "A_appendix · 完整稿件源码",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "8ac3f4391ebe659b2a91869adf36d1eab7bd148673a6ee2f48615fabb2d13457",
      "priority": 2,
      "public": true,
      "sha256": "8ac3f4391ebe659b2a91869adf36d1eab7bd148673a6ee2f48615fabb2d13457",
      "reviewed_sha256": "8ac3f4391ebe659b2a91869adf36d1eab7bd148673a6ee2f48615fabb2d13457",
      "bytes": 21114,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_A_APPENDIX/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/TEX_A_APPENDIX/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "MAIN_TEX",
      "path": "paper-rl/paper/main.tex",
      "title": "当前稿件主入口",
      "kind": "manuscript",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "bbb2b03a695ad7f47f6f0868dec76ead6362908c2cde25f057176a550736c344",
      "priority": 2,
      "public": true,
      "sha256": "bbb2b03a695ad7f47f6f0868dec76ead6362908c2cde25f057176a550736c344",
      "reviewed_sha256": "bbb2b03a695ad7f47f6f0868dec76ead6362908c2cde25f057176a550736c344",
      "bytes": 1202,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/MAIN_TEX/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/MAIN_TEX/source.tex",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "CURRENT_PDF",
      "path": "paper-rl/paper/main.pdf",
      "title": "当前PDF",
      "kind": "artifact",
      "observed_at": "2026-10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "expected_sha256": "096a8ff66cbbda0aa622f8511418bcc6691741f8b016f7c812a2366dd4ca9594",
      "priority": 2,
      "public": true,
      "sha256": "096a8ff66cbbda0aa622f8511418bcc6691741f8b016f7c812a2366dd4ca9594",
      "reviewed_sha256": "096a8ff66cbbda0aa622f8511418bcc6691741f8b016f7c812a2366dd4ca9594",
      "bytes": 297140,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CURRENT_PDF/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CURRENT_PDF/source.pdf",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986",
      "text_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/CURRENT_PDF/text.txt"
    },
    {
      "id": "RESUME_1105",
      "path": "paper-rl/results/resume_reconciliation_20261105.json",
      "title": "11-05只读接手与恢复材料复核",
      "kind": "receipt",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "priority": 1,
      "public": true,
      "expected_sha256": "84881c9babf5b3b229ff2aa313a831b9c17f45b60ac1fdf665394cd81b974223",
      "recorded_at": "2026-11-05T02:31:44+08:00",
      "sha256": "84881c9babf5b3b229ff2aa313a831b9c17f45b60ac1fdf665394cd81b974223",
      "reviewed_sha256": "84881c9babf5b3b229ff2aa313a831b9c17f45b60ac1fdf665394cd81b974223",
      "bytes": 4615,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RESUME_1105/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/RESUME_1105/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    },
    {
      "id": "WORKSPACE_CHECK_1105",
      "path": "paper-rl/results/workspace_sync_validation_20261105.json",
      "title": "11-05工作台同步验证与原失败",
      "kind": "receipt",
      "observed_at": "2026-11-05T02:32:58+08:00",
      "priority": 2,
      "public": true,
      "expected_sha256": "14277838156c4014cdadbd11898104ef3407207cc799835488ccfdb31a158653",
      "recorded_at": "2026-11-05T02:35:19+08:00",
      "sha256": "14277838156c4014cdadbd11898104ef3407207cc799835488ccfdb31a158653",
      "reviewed_sha256": "14277838156c4014cdadbd11898104ef3407207cc799835488ccfdb31a158653",
      "bytes": 1557,
      "read_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/WORKSPACE_CHECK_1105/",
      "raw_url": "https://sotalabs.cn/papers/_releases/v2-eb1a844c9b153986/rl/library/WORKSPACE_CHECK_1105/source.json",
      "source_changed_since_review": false,
      "release_version": "v2-eb1a844c9b153986"
    }
  ],
  "issues": [
    {
      "id": "RL-I001",
      "title": "恢复训练的证据闭环",
      "category": "执行",
      "priority": 1,
      "status": "阻塞",
      "description": "没有有效训练终点，就无法形成同预算策略比较。",
      "evidence": "11-05旧账本仍34次完成；step30 HF结构通过，父容器运行但gpu08未见匹配RL入口，GPU活动尚未归属。新ownership、恢复审定和release仍缺。[[RESUME_1105]][[RECOVERY_ARTIFACT]]",
      "attempts": [
        "参见关联证据与历史路线；本次导入没有新增科研尝试。"
      ],
      "unknowns": "当前可分配卡位/使用窗口及GPU作业归属尚未确认；容器重启不证明原run恢复。",
      "next_action": "确认当前资源与唯一调度负责人；按冻结协议审定外部终止后的恢复计划和精确checkpoint，绑定新preflight/release后才执行。",
      "done": "恢复release明确且数值/参数检查通过，run可追溯；不得以容器启动状态替代。",
      "dependencies": [],
      "first_discovered_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "updated_at": "2026-11-05T02:31:44+08:00",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "sources": [
        "R4",
        "RECOVERY_ARTIFACT",
        "RECOVERY_PROTOCOL",
        "RESUME_1105"
      ],
      "resolved_evidence": [],
      "closure_reason": null,
      "history": [
        {
          "at": "2026-11-05T02:31:44+08:00",
          "status": "阻塞",
          "note": "补充11-05只读现场及HF结构证据，不放行、不关闭恢复问题。",
          "event": "RL-20261105-RESUME-CHECK-01"
        }
      ],
      "journals": [
        "2026-09-28",
        "2026-10-01"
      ]
    },
    {
      "id": "RL-I002",
      "title": "逐槽位评测准入与unknown",
      "category": "证据",
      "priority": 2,
      "status": "待处理",
      "description": "影响固定分母和主成绩的可信性。",
      "evidence": "1000控制中仍有113未决、20部分观测零及5预期差异。[[R5]]",
      "attempts": [
        "参见关联证据与历史路线；本次导入没有新增科研尝试。"
      ],
      "unknowns": "具体缺口与现场状态仍需核对，不能从旧记录推断当前进程。",
      "next_action": "接入已完成归因与观测规则，保持原始attempt，不事后追认方便的分数。",
      "done": "正式评测manifest绑定环境、名单、模型与unknown规则，并有独立准入审查。",
      "dependencies": [],
      "first_discovered_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "updated_at": "2026-10-04T11:56:20+08:00",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R5"
      ],
      "resolved_evidence": [],
      "closure_reason": null,
      "history": [],
      "journals": [
        "2026-09-30"
      ]
    },
    {
      "id": "RL-I003",
      "title": "局部分配是否产生真实增益",
      "category": "研究",
      "priority": 3,
      "status": "待处理",
      "description": "决定这篇能否成为有效的方法论文。",
      "evidence": "非零信号已经观察到，M−B等策略效果仍未测量。[[R1]]",
      "attempts": [
        "参见关联证据与历史路线；本次导入没有新增科研尝试。"
      ],
      "unknowns": "具体缺口与现场状态仍需核对，不能从旧记录推断当前进程。",
      "next_action": "在前两项关闭后完成登记的O/B/M/E、种子和同栈评测。",
      "done": "主比较真实可复算，并报告无效/缺判分率与成本；无收益时保留负结论。",
      "dependencies": [],
      "first_discovered_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "updated_at": "2026-10-04T11:56:20+08:00",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R1"
      ],
      "resolved_evidence": [],
      "closure_reason": null,
      "history": [],
      "journals": []
    },
    {
      "id": "RL-I004",
      "title": "从协议稿转成效果稿",
      "category": "写作",
      "priority": 4,
      "status": "待处理",
      "description": "现稿不能承担尚未成立的性能主张。",
      "evidence": "submission_ready=false、effectiveness pending。[[R6]][[R8]]",
      "attempts": [
        "参见关联证据与历史路线；本次导入没有新增科研尝试。"
      ],
      "unknowns": "具体缺口与现场状态仍需核对，不能从旧记录推断当前进程。",
      "next_action": "真实效果判定后重写主结果、摘要和结论，重做最终论断/引用与改稿检查。",
      "done": "最终版本的每个效果与费用主张有证据，验收报告与PDF哈希绑定。",
      "dependencies": [],
      "first_discovered_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "updated_at": "2026-10-04T11:56:20+08:00",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R6",
        "R8"
      ],
      "resolved_evidence": [],
      "closure_reason": null,
      "history": [],
      "journals": [
        "2026-09-27",
        "2026-09-28"
      ]
    },
    {
      "id": "RL-I005",
      "title": "旧B80首步判分污染与熵归因",
      "category": "证据",
      "priority": 2,
      "status": "阻塞",
      "description": "影响主要M−B对照的有效性。",
      "evidence": "首步前B有203条地址池耗尽未判分；mask路径不能解释熵差，raw batch证据未保存。[[B_PREFLIGHT]][[ENTROPY_AUDIT]][[RAW_GAP]]",
      "attempts": [
        "已核对入口配置和response_mask路径。",
        "已检查保存checkpoint，不能恢复首步logits/mask/entropy。"
      ],
      "unknowns": "具体首步数值归因尚未由消费批次证据证明。",
      "next_action": "按清洁网络池与首步批次采集计划预检，未放行前不启动。",
      "done": "受控B首次消费批次收据、完整80步及同栈终评，而非旧reward比较。",
      "dependencies": [
        "RL-I001"
      ],
      "first_discovered_at": null,
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "updated_at": "2026-10-04T11:56:20+08:00",
      "observed_at": "2026-10-01T15:45:00+08:00",
      "sources": [
        "B_PREFLIGHT",
        "ENTROPY_AUDIT",
        "RAW_GAP",
        "B_PLAN"
      ],
      "resolved_evidence": [],
      "closure_reason": null,
      "history": [],
      "journals": [
        "2026-10-01"
      ]
    }
  ],
  "tasks": [
    {
      "id": "RL-T001",
      "title": "先关闭恢复与准入缺口",
      "status": "阻塞",
      "action": "已有11-05只读核对；先确认当前资源归属和调度责任，再审定终止后恢复的精确checkpoint与新release。不得仅凭父容器running重启。",
      "done": "产出恢复和评测manifest；不满足就停在核查，不能进入主比较。",
      "dependencies": [],
      "issues": [
        "RL-I001"
      ],
      "plan_kind": "网页整理的下一步；执行前按原研究协议核对",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "sources": [
        "RESUME_1105",
        "RECOVERY_PROTOCOL",
        "RECOVERY_ARTIFACT"
      ],
      "updated_at": "2026-11-05T02:31:44+08:00"
    },
    {
      "id": "RL-T002",
      "title": "完成登记的对照与终评",
      "status": "待处理",
      "action": "依赖前一步，按同预算和同栈完成B/M及O/E计划，保持holdout规则。",
      "done": "产出每臂每种子效果、失败比例、成本；M−B无收益则不宣称方法有效。",
      "dependencies": [
        "RL-T001"
      ],
      "issues": [
        "RL-I002"
      ],
      "plan_kind": "网页整理的下一步；执行前按原研究协议核对",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28"
    },
    {
      "id": "RL-T003",
      "title": "按结果重写并验收",
      "status": "待处理",
      "action": "用真实比较决定保留哪些贡献，再更新全文、图表、引用与最终PDF。",
      "done": "成立则写效果稿；不成立则保留结果并调整方向，不能用pilot替代。",
      "dependencies": [
        "RL-T002"
      ],
      "issues": [
        "RL-I003"
      ],
      "plan_kind": "网页整理的下一步；执行前按原研究协议核对",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28"
    }
  ],
  "experiments": [
    {
      "id": "RL-E001",
      "title": "非零删除信号",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "92/244=37.7%；其中失败原轨迹 17/111 有非零差（12 正、5 负）。",
      "interpretation": "支持信号可用性；不能证明归属准确或学到更好的策略。[[R1]]",
      "technical_status": "已记录",
      "scientific_status": "限于来源范围",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R1"
      ],
      "issues": [
        "RL-I003"
      ]
    },
    {
      "id": "RL-E002",
      "title": "辅助项 cap",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "92 个非零差中 83 次触发 cap（90.2%）。",
      "interpretation": "多数非零样本进入 sign/support 区域，需要分析优化几何；不把 D 幅度当成原样保留。[[R1]]",
      "technical_status": "已记录",
      "scientific_status": "限于来源范围",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R1"
      ],
      "issues": [
        "RL-I003"
      ]
    },
    {
      "id": "RL-E003",
      "title": "原产物复现 / 服务重复",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "120 次跨机器 factual replay 的报告标签一致；244 次干预重复报告一致。",
      "interpretation": "支持这组记录的复现，未覆盖所有生产环境与模型终评。[[R1]]",
      "technical_status": "已记录",
      "scientific_status": "限于来源范围",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R1"
      ],
      "issues": [
        "RL-I003"
      ]
    },
    {
      "id": "RL-E004",
      "title": "共同环境最终覆盖",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "500 题/1000 控制；raw 符合预期 995，strict 符合预期 862，报告二元 887，未决 113。",
      "interpretation": "这些是控制级覆盖；其中 20 个部分观测零、5 个预期差异仍保留。不是模型分数。[[R5]]",
      "technical_status": "已记录",
      "scientific_status": "限于来源范围",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R5"
      ],
      "issues": [
        "RL-I002"
      ]
    },
    {
      "id": "RL-E005",
      "title": "O42-retry1 恢复记录",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "截至 09-28 完成训练调用 34/80；父容器 exit137，OOMKilled=false。",
      "interpretation": "终止归因仍未决；没有自动重启和恢复证明，不显示为实时训练中。[[R4]]",
      "technical_status": "已记录",
      "scientific_status": "限于来源范围",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R4"
      ],
      "issues": [
        "RL-I001"
      ]
    },
    {
      "id": "RL-E006",
      "title": "M−B / M−O / M−E",
      "purpose": "核对当前主张的证据范围；详见原文设置。",
      "setup": "pilot 原有 40 任务×8 条，正式训练准入为 149 任务，相关开发子集为 39 任务。主评测官方500和封存holdout100保留固定分母；开发机制 pilot 不能冒充 holdout 效果。[[R2]][[R6]]",
      "result": "未测量；留出策略效果、种子比较和总训练成本 Pending。",
      "interpretation": "主方法的实证贡献尚未完成。[[R1]][[R6]]",
      "technical_status": "未测量",
      "scientific_status": "待验证",
      "cost": "按来源的输入、输出、服务时间与GPU小时口径阅读，未知不补零。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "sources": [
        "R1",
        "R6"
      ],
      "issues": [
        "RL-I003",
        "RL-I004"
      ]
    },
    {
      "id": "RL-E008",
      "title": "终止后step30恢复材料",
      "purpose": "核对恢复/对照有效性",
      "setup": "只读源文件、checkpoint与保留日志，不启动GPU任务。",
      "result": "材料存在，但launch_ready=false；不证明训练已恢复。",
      "interpretation": "技术审计，不是模型效果。",
      "technical_status": "审计完成",
      "scientific_status": "待准入/受控重跑",
      "cost": "只读审计；原训练成本不由本次检查推算。",
      "observed_at": "2026-10-01",
      "sources": [
        "RECOVERY_ARTIFACT"
      ],
      "issues": [
        "RL-I001"
      ]
    },
    {
      "id": "RL-E009",
      "title": "B首步判分与熵审计",
      "purpose": "核对恢复/对照有效性",
      "setup": "只读源文件、checkpoint与保留日志，不启动GPU任务。",
      "result": "203条未判分均为地址池耗尽，旧B80主比较受污染。",
      "interpretation": "技术审计，不是模型效果。",
      "technical_status": "审计完成",
      "scientific_status": "待准入/受控重跑",
      "cost": "只读审计；原训练成本不由本次检查推算。",
      "observed_at": "2026-10-01",
      "sources": [
        "B_PREFLIGHT",
        "ENTROPY_AUDIT",
        "RAW_GAP"
      ],
      "issues": [
        "RL-I005"
      ]
    },
    {
      "id": "RL-E010",
      "title": "11-05恢复状态与HF结构只读复核",
      "purpose": "区分历史运行记录、当前资源活动与有效训练恢复证据。",
      "setup": "固定旧O42-retry1目录；读取消费账本和step30索引/头部/文件长度；两机非同步瞬时资源采样，0模型调用、0新GPU作业。",
      "result": "34/80调用记录未续进；step30四分片427张量结构通过；两机父容器running，但未建立旧run恢复证据。",
      "interpretation": "只证明所读材料与现场快照；不证明优化器恢复、权重载入或模型效果。[[RESUME_1105]]",
      "technical_status": "只读核对完成",
      "scientific_status": "训练恢复未放行",
      "cost": "只读检查；不估算此前训练总成本。",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "sources": [
        "RESUME_1105"
      ],
      "issues": [
        "RL-I001"
      ]
    }
  ],
  "decisions": [
    {
      "id": "RL-D001",
      "title": "旧污染与无效训练",
      "question": "这条路线是否仍适合承担当前研究主张？",
      "alternatives": [
        "继续作为主方法",
        "保留为诊断/基线",
        "关闭或等待准入"
      ],
      "observation": "旧M历史存在污染；旧B/O也不能简单当成当前干净对照。",
      "choice": "原日志和存点归档，不能拿旧在线reward填新的方法效果。[[R7]]",
      "reason": "以原记录中的证据与门槛为依据，不另造因果解释。",
      "expected": "仅按明确贡献范围继续；未执行计划不能当结果。",
      "actual": "旧M历史存在污染；旧B/O也不能简单当成当前干净对照。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "sources": [
        "R7"
      ],
      "issues": []
    },
    {
      "id": "RL-D002",
      "title": "评分适配器恢复",
      "question": "这条路线是否仍适合承担当前研究主张？",
      "alternatives": [
        "继续作为主方法",
        "保留为诊断/基线",
        "关闭或等待准入"
      ],
      "observation": "早先存在无完整grading report的零分路径。",
      "choice": "修复后需完整判分证据；缺反馈可能减少模型安装失败的负反馈，按臂报告。[[R3]]",
      "reason": "以原记录中的证据与门槛为依据，不另造因果解释。",
      "expected": "仅按明确贡献范围继续；未执行计划不能当结果。",
      "actual": "早先存在无完整grading report的零分路径。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "sources": [
        "R3"
      ],
      "issues": []
    },
    {
      "id": "RL-D003",
      "title": "O42-retry1",
      "question": "这条路线是否仍适合承担当前研究主张？",
      "alternatives": [
        "继续作为主方法",
        "保留为诊断/基线",
        "关闭或等待准入"
      ],
      "observation": "有数值/存点检查，34次完成调用后外部终止。",
      "choice": "恢复需绑定准确checkpoint与终止证据，容器重启不等于研究run恢复。[[R4]][[R7]]",
      "reason": "以原记录中的证据与门槛为依据，不另造因果解释。",
      "expected": "仅按明确贡献范围继续；未执行计划不能当结果。",
      "actual": "有数值/存点检查，34次完成调用后外部终止。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "sources": [
        "R4",
        "R7"
      ],
      "issues": [
        "RL-I001"
      ]
    },
    {
      "id": "RL-D004",
      "title": "共同环境覆盖",
      "question": "这条路线是否仍适合承担当前研究主张？",
      "alternatives": [
        "继续作为主方法",
        "保留为诊断/基线",
        "关闭或等待准入"
      ],
      "observation": "全部500题控制执行完毕，聚合后续纠正。",
      "choice": "保留unknown与环境差异，完成覆盖只是准入工作的输入。[[R5]]",
      "reason": "以原记录中的证据与门槛为依据，不另造因果解释。",
      "expected": "仅按明确贡献范围继续；未执行计划不能当结果。",
      "actual": "全部500题控制执行完毕，聚合后续纠正。",
      "observed_at": "2026-10-01 汇总更正；训练记录止于 09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "sources": [
        "R5"
      ],
      "issues": [
        "RL-I002"
      ]
    },
    {
      "id": "RL-D005",
      "title": "不把GPU繁忙或容器运行当作O恢复",
      "question": "是否可以沿用旧运行状态直接继续训练？",
      "alternatives": [
        "依据旧交接重启",
        "先核对归属和新恢复release"
      ],
      "observation": "旧账本停止在34次完成，父容器后来运行；当前卡忙但未归属到该run，旧launch_ready=false未被新证据替代。",
      "choice": "保持恢复阻塞；不启停任务，不应用已存在的旧补丁。[[RESUME_1105]]",
      "reason": "冻结恢复协议要求精确checkpoint、终止证据、ownership和新release；当前材料不足。",
      "expected": "由可审计的新证据决定恢复，避免抢占其他任务或混淆模型结果。",
      "actual": "完成只读复核，未启动/停止任何作业，未改科学协议。",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "recorded_at": "2026-11-05T02:25:51+08:00",
      "sources": [
        "RESUME_1105",
        "RECOVERY_PROTOCOL"
      ],
      "issues": [
        "RL-I001"
      ]
    },
    {
      "id": "RL-D006",
      "title": "工作台观察时间校验跟随源记录",
      "question": "如何接纳新观察同时保留旧事件？",
      "alternatives": [
        "固定旧日期常量",
        "验证导出与源记录一致并保持历史不可变"
      ],
      "observation": "合法11-05更新触发固定10-01断言失败，非来源漂移或数据缺失。[[WORKSPACE_CHECK_1105]]",
      "choice": "改为全项目导出/源记录时间一致性检查，并增加历史事件不变回归。",
      "reason": "验证工作台不变量而非将某次历史快照固化成永久日期。",
      "expected": "新日期可以同步，历史内容和来源检查继续受保护。",
      "actual": "14项测试通过；科研准入未改变。",
      "observed_at": "2026-11-05T02:32:58+08:00",
      "recorded_at": "2026-11-05T02:32:58+08:00",
      "sources": [
        "WORKSPACE_CHECK_1105"
      ],
      "issues": []
    }
  ],
  "events": [
    {
      "id": "RL-H001",
      "date": "2026-09-20",
      "title": "旧运行完整性复盘",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-09-20—24",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "旧运行完整性复盘"
        ],
        "完成的工作": [
          "污染、评分与资源故障使历史结果不能直接进入主比较。[[R7]]"
        ],
        "新证据与结果": [
          "污染、评分与资源故障使历史结果不能直接进入主比较。[[R7]]"
        ],
        "判断与决策变化": [
          "补录仅归纳原记录；详细决定与历史结果见关联原文。"
        ],
        "问题变化": [
          "关联问题用于追踪背景；这里不推断当天已解决。"
        ],
        "论文变化": [
          "具体稿件版本见论文栏目；本条历史补录不推断当天改稿。"
        ],
        "下一步与交接": [
          "这是历史记录，当前行动以最新概览和接手包为准。"
        ]
      },
      "sources": [
        "R7"
      ],
      "issues": [],
      "experiments": [],
      "decisions": [
        "RL-D001",
        "RL-D003"
      ],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-H002",
      "date": "2026-09-27",
      "title": "方法/协议稿建立",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-09-27",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "方法/协议稿建立"
        ],
        "完成的工作": [
          "固定最终产物删除和O/B/M/E比较，pilot提供信号观察。[[R2]][[R6]]"
        ],
        "新证据与结果": [
          "固定最终产物删除和O/B/M/E比较，pilot提供信号观察。[[R2]][[R6]]"
        ],
        "判断与决策变化": [
          "补录仅归纳原记录；详细决定与历史结果见关联原文。"
        ],
        "问题变化": [
          "关联问题用于追踪背景；这里不推断当天已解决。"
        ],
        "论文变化": [
          "具体稿件版本见论文栏目；本条历史补录不推断当天改稿。"
        ],
        "下一步与交接": [
          "这是历史记录，当前行动以最新概览和接手包为准。"
        ]
      },
      "sources": [
        "R2",
        "R6"
      ],
      "issues": [
        "RL-I004"
      ],
      "experiments": [
        "RL-E006"
      ],
      "decisions": [],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-H003",
      "date": "2026-09-28",
      "title": "恢复稿补充限制",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "恢复稿补充限制"
        ],
        "完成的工作": [
          "披露缺反馈和环境问题，保留draft。[[R8]]"
        ],
        "新证据与结果": [
          "披露缺反馈和环境问题，保留draft。[[R8]]"
        ],
        "判断与决策变化": [
          "补录仅归纳原记录；详细决定与历史结果见关联原文。"
        ],
        "问题变化": [
          "关联问题用于追踪背景；这里不推断当天已解决。"
        ],
        "论文变化": [
          "具体稿件版本见论文栏目；本条历史补录不推断当天改稿。"
        ],
        "下一步与交接": [
          "这是历史记录，当前行动以最新概览和接手包为准。"
        ]
      },
      "sources": [
        "R8"
      ],
      "issues": [
        "RL-I004"
      ],
      "experiments": [],
      "decisions": [],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-H004",
      "date": "2026-09-28",
      "title": "O恢复外部终止",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-09-28",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "O恢复外部终止"
        ],
        "完成的工作": [
          "34次完成调用；终止不等于方法失败或恢复成功。[[R4]]"
        ],
        "新证据与结果": [
          "34次完成调用；终止不等于方法失败或恢复成功。[[R4]]"
        ],
        "判断与决策变化": [
          "补录仅归纳原记录；详细决定与历史结果见关联原文。"
        ],
        "问题变化": [
          "关联问题用于追踪背景；这里不推断当天已解决。"
        ],
        "论文变化": [
          "具体稿件版本见论文栏目；本条历史补录不推断当天改稿。"
        ],
        "下一步与交接": [
          "这是历史记录，当前行动以最新概览和接手包为准。"
        ]
      },
      "sources": [
        "R4"
      ],
      "issues": [
        "RL-I001"
      ],
      "experiments": [
        "RL-E005"
      ],
      "decisions": [
        "RL-D003"
      ],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-H005",
      "date": "2026-09-30",
      "title": "500控制覆盖并纠错",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-09-30—10-01",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "500控制覆盖并纠错"
        ],
        "完成的工作": [
          "控制执行覆盖完成，最终聚合重算，未决继续保留。[[R5]][[R7]]"
        ],
        "新证据与结果": [
          "控制执行覆盖完成，最终聚合重算，未决继续保留。[[R5]][[R7]]"
        ],
        "判断与决策变化": [
          "补录仅归纳原记录；详细决定与历史结果见关联原文。"
        ],
        "问题变化": [
          "关联问题用于追踪背景；这里不推断当天已解决。"
        ],
        "论文变化": [
          "具体稿件版本见论文栏目；本条历史补录不推断当天改稿。"
        ],
        "下一步与交接": [
          "这是历史记录，当前行动以最新概览和接手包为准。"
        ]
      },
      "sources": [
        "R5",
        "R7"
      ],
      "issues": [
        "RL-I002"
      ],
      "experiments": [
        "RL-E004"
      ],
      "decisions": [
        "RL-D001",
        "RL-D003",
        "RL-D004"
      ],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-ORIGINAL-2026-09-07",
      "date": "2026-09-07",
      "title": "09-07原始日报",
      "kind": "原始日报",
      "occurred_at": null,
      "observed_at": "2026-09-07",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "原始日报完整保留；混合时点按正文阅读。"
        ],
        "完成的工作": [
          "查看原始日报正文。"
        ],
        "新证据与结果": [
          "所有数字保留原条件，不能用历史运行状态覆盖当前状态。"
        ],
        "判断与决策变化": [
          "原文保留当时的判断与后续补充。"
        ],
        "问题变化": [
          "参见当日原文，不自动追认已解决。"
        ],
        "论文变化": [
          "参见原文稿件记录。"
        ],
        "下一步与交接": [
          "历史下一步不是当前执行指令。"
        ]
      },
      "sources": [
        "DIARY_0907"
      ],
      "issues": [],
      "experiments": [],
      "decisions": [],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-ORIGINAL-2026-09-08",
      "date": "2026-09-08",
      "title": "09-08原始日报",
      "kind": "原始日报",
      "occurred_at": null,
      "observed_at": "2026-09-08",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "原始日报完整保留；混合时点按正文阅读。"
        ],
        "完成的工作": [
          "查看原始日报正文。"
        ],
        "新证据与结果": [
          "所有数字保留原条件，不能用历史运行状态覆盖当前状态。"
        ],
        "判断与决策变化": [
          "原文保留当时的判断与后续补充。"
        ],
        "问题变化": [
          "参见当日原文，不自动追认已解决。"
        ],
        "论文变化": [
          "参见原文稿件记录。"
        ],
        "下一步与交接": [
          "历史下一步不是当前执行指令。"
        ]
      },
      "sources": [
        "DIARY_0908"
      ],
      "issues": [],
      "experiments": [],
      "decisions": [],
      "tasks": [],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-H006",
      "date": "2026-10-01",
      "title": "恢复材料与B主对照最新审计",
      "kind": "补录",
      "occurred_at": null,
      "observed_at": "2026-10-01T15:45:00+08:00",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "O恢复与B重跑仍未放行；旧B80不能作为干净M−B主对照。"
        ],
        "完成的工作": [
          "核对step30材料、入口配置、熵mask与首更新前日志。"
        ],
        "新证据与结果": [
          "B首熵前203条未判分均为地址池耗尽；现存存点不能重建首步熵批次。"
        ],
        "判断与决策变化": [
          "排除broadcast改变entropy mask的解释；保持B数值归因未决。"
        ],
        "问题变化": [
          "恢复release与B受控重跑需先关闭准入缺口。"
        ],
        "论文变化": [
          "主效果未测；不能用旧B/M在线reward建立方法胜负。"
        ],
        "下一步与交接": [
          "现场确认资源、先完成新恢复release和首步批次采集。"
        ]
      },
      "sources": [
        "RECOVERY_ARTIFACT",
        "B_PREFLIGHT",
        "ENTROPY_AUDIT",
        "RAW_GAP",
        "B_PLAN"
      ],
      "issues": [
        "RL-I001",
        "RL-I005"
      ],
      "experiments": [
        "RL-E008",
        "RL-E009"
      ],
      "decisions": [],
      "tasks": [
        "RL-T001"
      ],
      "artifacts": [],
      "corrects": null
    },
    {
      "id": "RL-SITE-V2",
      "date": "2026-10-04",
      "title": "研究工作台资料同步",
      "kind": "工作记录",
      "occurred_at": "2026-10-04T11:56:20+08:00",
      "observed_at": "2026-10-01T15:45:00+08:00",
      "recorded_at": "2026-10-04T11:56:20+08:00",
      "fields": {
        "当天结论": [
          "研究工作台上线资料整理；没有新增科研实验。"
        ],
        "完成的工作": [
          "建立日报、问题、实验、稿件与完整资料的关联入口。"
        ],
        "新证据与结果": [
          "公开现有来源，所有科研数字沿用各自观察时间。"
        ],
        "判断与决策变化": [
          "科研结论不因网站上线改变；网页用于恢复上下文。"
        ],
        "问题变化": [
          "当前问题按既有证据收录，没有自动关闭。"
        ],
        "论文变化": [
          "保留现有研究说明与PDF版本；未改写论文。"
        ],
        "下一步与交接": [
          "后续参与工作的AI用同步入口更新记录、验证并发布；仅阅读者无需写入。"
        ]
      },
      "sources": [],
      "issues": [],
      "experiments": [],
      "decisions": [],
      "tasks": [],
      "artifacts": [
        "CURRENT_PDF"
      ],
      "corrects": null
    },
    {
      "id": "RL-20261105-RESUME-CHECK-01",
      "date": "2026-11-05",
      "title": "重新接手：旧O账本未续进，恢复材料存在但未放行",
      "kind": "工作记录",
      "occurred_at": "2026-11-05T02:25:51+08:00",
      "observed_at": "2026-11-05T02:25:51+08:00",
      "fields": {
        "当天结论": [
          "旧O42-retry1账本仍为34/80次完成调用；GPU繁忙和父容器运行不证明该run恢复，launch_ready仍为false。[[RESUME_1105]]"
        ],
        "完成的工作": [
          "核对本地RL记录、线上固定版本交接、终止与恢复收据、消费账本、两机只读资源/容器状态和step30 HF结构。没有启动、停止或清理任务。"
        ],
        "新证据与结果": [
          "账本34次training_call_completed、35次selected/advantages_prepared；step30 HF四分片、427张量、17,907,606,528 payload bytes结构检查通过，不是加载/恢复验证。两机卡均有活动，归属未确立。[[RESUME_1105]]"
        ],
        "判断与决策变化": [
          "不再沿用9月早期的87/120活跃状态；也不将10月旧收据当作11月运行证明。9月恢复查找错误的最小修复已存在于当前源码，不重复应用旧补丁。"
        ],
        "问题变化": [
          "RL-I001保持阻塞，新增本次现场与材料核对证据；RL-T001转为明确阻塞，仍需资源归属、恢复决策和新release。其余科学问题未关闭。"
        ],
        "论文变化": [
          "未修改稿件或方法协议，未读取封存成绩；主效果仍未测。发布连接器初检传输失败，构建或打包不计为发布成功。"
        ],
        "下一步与交接": [
          "先确认当前可分配GPU与调度负责人，再按冻结恢复协议审定checkpoint及新release；工作台使用既有同步/构建/发布流程，发布失败保留线上前驱。"
        ]
      },
      "sources": [
        "RESUME_1105",
        "R4",
        "RECOVERY_ARTIFACT",
        "RECOVERY_PROTOCOL"
      ],
      "issues": [
        "RL-I001"
      ],
      "experiments": [
        "RL-E010"
      ],
      "decisions": [
        "RL-D005"
      ],
      "tasks": [
        "RL-T001"
      ],
      "artifacts": [
        "RESUME_1105"
      ],
      "corrects": null,
      "recorded_at": "2026-11-05T02:31:44+08:00"
    },
    {
      "id": "RL-20261105-WORKSPACE-CHECK-02",
      "date": "2026-11-05",
      "title": "工作台日期回归修正与发布连接恢复",
      "kind": "工作记录",
      "occurred_at": "2026-11-05T02:32:58+08:00",
      "observed_at": "2026-11-05T02:32:58+08:00",
      "fields": {
        "当天结论": [
          "工作台14项单测通过；这是文档基础设施验证，不是训练或模型效果。[[WORKSPACE_CHECK_1105]]"
        ],
        "完成的工作": [
          "将写死10月1日的断言替换为导出/源记录时间一致性校验，保留冲突检查，并加入历史事件不变回归。"
        ],
        "新证据与结果": [
          "初次13项测试中1项失败及原因保留；修正后14/14通过，JavaScript语法检查通过。发布连接器重试恢复，尚需完整stage/activate/verify。"
        ],
        "判断与决策变化": [
          "合法新观察不能被旧日期常量阻断；仍由源记录决定导出的时间，不用回改历史日期获得通过。"
        ],
        "问题变化": [
          "科研问题RL-I001等不因网站测试通过而关闭；本轮没有新增训练放行。"
        ],
        "论文变化": [
          "论文、科学协议和历史失败未改；不静默确认任何来源漂移。"
        ],
        "下一步与交接": [
          "构建完整版本并核对线上前驱，按既有连接器流程原子发布；实际发布结果只认后续验收收据。"
        ]
      },
      "sources": [
        "WORKSPACE_CHECK_1105"
      ],
      "issues": [],
      "experiments": [],
      "decisions": [
        "RL-D006"
      ],
      "tasks": [],
      "artifacts": [
        "WORKSPACE_CHECK_1105"
      ],
      "corrects": null,
      "recorded_at": "2026-11-05T02:35:19+08:00"
    }
  ],
  "restricted": [
    {
      "id": "SEALED_TEST",
      "title": "封存测试与任务级标签",
      "path": "原项目的holdout/test材料",
      "availability": "未公开",
      "access": "在原环境按冻结协议与授权访问；发布流程不读取。"
    },
    {
      "id": "LARGE_ARTIFACTS",
      "title": "模型权重与完整轨迹",
      "path": "原项目模型/run目录",
      "availability": "体量过大，未发布",
      "access": "通过原项目文件清单、run身份及授权环境获取。"
    },
    {
      "id": "ACCOUNT_CREDENTIALS",
      "title": "账号凭据",
      "path": null,
      "availability": "不公开",
      "access": "使用已有连接或原环境授权，不写入接手文本。"
    }
  ],
  "manuscript": {
    "title": "Counterfactual Edit Credit for Long-Horizon Software Agents",
    "revision": "2026-10-01",
    "pdf_document": "CURRENT_PDF",
    "rows": [
      [
        "摘要 / 引言",
        "已有草稿",
        "写清机制与信号观察，禁止预告未存在的效果增益。[[R1]][[R6]]"
      ],
      [
        "相关工作 / 方法",
        "已有草稿",
        "定义块选择、恢复、效用、B/M/E及优化限制。[[R2]][[R3]]"
      ],
      [
        "实验设计",
        "已有草稿",
        "计划与已执行分开；终点、训练题与留出划分需按真实release更新。[[R2]]"
      ],
      [
        "策略主结果 / 总成本",
        "待新证据重写",
        "M−B等及总体成本仍未测量，主表明确Pending。[[R1]]"
      ],
      [
        "机制结果",
        "已有草稿",
        "pilot、cap、复现与服务时间有来源；不升格为能力提升。[[R1]]"
      ],
      [
        "限制 / 附录",
        "已核对",
        "已包含干预estimand、优化非等价与判分恢复限制。[[R3]][[R8]]"
      ],
      [
        "图表 / 引用 / 审查",
        "已有草稿",
        "示意图与实测表分开；旧审查只覆盖指定范围，最终效果稿需再次验收。[[R6]][[R8]]"
      ],
      [
        "PDF",
        "已核对",
        "当前15页产物存在；历史恢复稿记录正文9页，页数不证明实验完成。[[R8]]"
      ]
    ],
    "notes": [
      "现稿有方法、实验协议、机制pilot、限制和附录；数值审查与写作循环在既有记录中属于same-family/provisional范围。稿件验收约定明确这是draft，不等于方法有效或投稿就绪。当前PDF已有10-01产物，部分审查记录早于它。[[R6]][[R8]]",
      "最终论文更新应在真实结果后重新建立验收约定，补齐结果与成本，再执行完整改稿与编译流程。当前保留draft状态；局部审查critical/major清零也不能替代终评。[[R6]][[R8]]"
    ]
  },
  "background_file": "rl.json",
  "ingested_records": {
    "RL-20261105-RESUME-CHECK-01": "6fdf2c6a44118d23156400c375ef78e33842b49ebc96390979d22b16f6f6f33c",
    "RL-20261105-WORKSPACE-CHECK-02": "e7ce8b9194c31ab465aef4df4499e1a1ef077f9b8721d622e890aa9afbdd2c58"
  },
  "release_version": "v2-eb1a844c9b153986",
  "snapshot_recorded_at": "2026-11-05T02:35:19+08:00",
  "source_drift": []
}
