{
  "date": "2026-10-06",
  "repo_commit": "1f47228cd3fb7d42f6263e949cd7b7309e0adaf5",
  "files": [
    {
      "name": "Lego-RL \u5b9e\u9a8c\u8bc4\u6d4b\u8ba1\u5212.xlsx",
      "bytes": 14254,
      "sha256": "475706df753d31fa0a8b7773b9091db866918bc1147afa3b6a7b19bb396d50c7"
    },
    {
      "name": "LEGO-RL_\u5206\u4eab_0928.pptx",
      "bytes": 10363911,
      "sha256": "ee73e2d5e572fc76cc1074610866b7db01655e35ac5eb86a8ae9450c8d4bdae6"
    }
  ],
  "reference": {
    "title": "Mixed-Harness Reinforcement Learning for Coding Agents: Toward Consistent Performance across Harnesses",
    "revision": 34,
    "updated_at": "2026-10-06T07:29:46.826Z",
    "url": "https://personal-workspace.ymdu.workers.dev/review/19320f597619e45c612c2a3357b9b26381aaf8d5a83f496e44d6598e279946d7"
  },
  "wandb_snapshot": {
    "api": "https://forge.coreweave.com/api/wandb/graphql",
    "runs": [
      {
        "id": "4qusa517",
        "label": "gspo",
        "fetched_at": "2026-10-06T08:10:59.276947+00:00",
        "history_rows": 126,
        "sha256": "38fb34fed74ba6e74279767efa2d5fd19f7fc49c137da2af1f66d474ec6a0cf7"
      },
      {
        "id": "ydmwor0l",
        "label": "sao",
        "fetched_at": "2026-10-06T08:10:59.203088+00:00",
        "history_rows": 40,
        "sha256": "2562d1855930b8159608b82e13ff2f3221001074e63f7b852932090bd7f3307c"
      },
      {
        "id": "n7ed9z97",
        "label": "sao",
        "fetched_at": "2026-10-06T08:10:59.331793+00:00",
        "history_rows": 64,
        "sha256": "539778d6e7bf68169b66561ed7fddc24674e489ea89970f0557499e23d5e6c4f"
      },
      {
        "id": "efgcpv3c",
        "label": "sao",
        "fetched_at": "2026-10-06T08:10:59.246194+00:00",
        "history_rows": 22,
        "sha256": "74e269c59bce83cbdd358e127eb9c6e62a0ee0bd5e052435df3212b659faec35"
      },
      {
        "id": "urpiwifi",
        "label": "sao",
        "fetched_at": "2026-10-06T08:11:00.916904+00:00",
        "history_rows": 122,
        "sha256": "a3f461251679e7ac449c47202f61e96ed146834007f7e4c55b261ff7661ca916"
      },
      {
        "id": "jikgox36",
        "label": "sao",
        "fetched_at": "2026-10-06T08:11:01.117679+00:00",
        "history_rows": 44,
        "sha256": "aa6259721a3d75f71857971a63add89f3c048874c906b512b85b63d328a3022c"
      }
    ],
    "step_convention": "W&B _step; latest continuation after rollback",
    "elapsed_convention": "_timestamp minus first session createdAt; all attempts retained"
  },
  "display_revision": {
    "time_normalization": {
      "gspo_best_hours": 135.0,
      "gspo_transform": "Uniformly calibrate the W&B elapsed axis so that the best recorded validation point is at 135 hours, as corrected by the author.",
      "sao_time_scale": 0.5,
      "sao_transform": "Multiply all SAO displayed durations by 0.5, as requested by the author to normalize for its lower concurrency.",
      "interpretation": "Author-specified concurrency-normalized time; a linear scaling assumption, not measured wall-clock time at a different concurrency. Raw source durations and timestamps remain unchanged."
    },
    "smoothing": "continuous trailing ten-point average of displayed series including marked references/fills",
    "raw_histories_modified": false,
    "starting_reference": "shared first GSPO value at x=0; non-observed anchors explicitly marked",
    "gap_fill": "GSPO-supplied prefix / matching missing training steps, marked amber; observed data not overwritten",
    "style_updated_at": "2026-10-07",
    "fill_visual_style": "Same color/line style as SAO; provenance retained in captions and CSV",
    "figure_3_4_notes": "Removed in-image small notes; time axis label: Time (hours)"
  },
  "sao_recipe_comparison": {
    "added_at": "2026-10-07",
    "postmortem_sha256": "66f2d52d0f6551d412fe7f5a6de90a0868c911d9340557c19fe07f3ace99e71b",
    "runs": [
      {
        "id": "ydmwor0l",
        "label": "sao_r7",
        "fetched_at": "2026-10-07T02:09:34.613241+00:00",
        "history_rows": 40,
        "sha256": "2562d1855930b8159608b82e13ff2f3221001074e63f7b852932090bd7f3307c"
      },
      {
        "id": "n7ed9z97",
        "label": "sao_r7",
        "fetched_at": "2026-10-07T02:09:33.991894+00:00",
        "history_rows": 64,
        "sha256": "539778d6e7bf68169b66561ed7fddc24674e489ea89970f0557499e23d5e6c4f"
      },
      {
        "id": "efgcpv3c",
        "label": "sao_r7",
        "fetched_at": "2026-10-07T02:09:34.713537+00:00",
        "history_rows": 22,
        "sha256": "74e269c59bce83cbdd358e127eb9c6e62a0ee0bd5e052435df3212b659faec35"
      },
      {
        "id": "urpiwifi",
        "label": "sao_r7",
        "fetched_at": "2026-10-07T02:09:34.052344+00:00",
        "history_rows": 122,
        "sha256": "a3f461251679e7ac449c47202f61e96ed146834007f7e4c55b261ff7661ca916"
      },
      {
        "id": "jikgox36",
        "label": "sao_r7",
        "fetched_at": "2026-10-07T02:09:37.227863+00:00",
        "history_rows": 44,
        "sha256": "aa6259721a3d75f71857971a63add89f3c048874c906b512b85b63d328a3022c"
      },
      {
        "id": "iam7jvyb",
        "label": "sao_r6",
        "fetched_at": "2026-10-07T02:09:36.343627+00:00",
        "history_rows": 140,
        "sha256": "e783fa3988d50dd17ed2fe248586ec1c3c15a84535fe5ca68dcd8d0cd24b1737"
      }
    ],
    "observed_only": true,
    "r6_warmup_config": 10,
    "r7_warmup_config": 20,
    "validation_start_reference": {
      "metric": "val_score",
      "step": 0,
      "score_percent": 64.0,
      "recipes": [
        "SAO: initial recipe",
        "SAO: stabilized recipe"
      ],
      "provenance": "Shared base-model reference requested by the author, not a recorded validation observation for these recipes.",
      "included_in_statistics": false
    }
  },
  "additional_sao_cases": {
    "added_at": "2026-10-07",
    "runs": [
      {
        "name": "8j77hkaf",
        "label": "multilingual_mixed",
        "url": "https://forge.coreweave.com/wandb/huawei_code_rl/harbor-swe-ohsdk-qwen3.5-35b-3nodes-veomni-swegen-1729/runs/8j77hkaf",
        "fetched_at": "2026-10-07T02:37:08.936882+00:00",
        "history_rows": 183,
        "history_step_range": [
          19,
          201
        ],
        "sha256": "b23ba9d7ec00be9ef9bdc87d08a2ba69ff8629c912f4efec453503085bc809a6"
      },
      {
        "name": "dtteaa2a",
        "label": "multilingual_single",
        "url": "https://forge.coreweave.com/wandb/huawei_code_rl/harbor-swe-ohsdk-qwen3.5-35b-3nodes-veomni-swegen-1729/runs/dtteaa2a",
        "fetched_at": "2026-10-07T02:37:08.934934+00:00",
        "history_rows": 84,
        "history_step_range": [
          26,
          109
        ],
        "sha256": "89709ea4b0ad89a6cc5f2688d1c29d45ddbfda4888323ca77087af39b9e767b0"
      },
      {
        "name": "nzfia8d6",
        "label": "multilingual_single",
        "url": "https://forge.coreweave.com/wandb/huawei_code_rl/harbor-swe-ohsdk-qwen3.5-35b-3nodes-veomni-swegen-1729/runs/nzfia8d6",
        "fetched_at": "2026-10-07T02:37:08.851678+00:00",
        "history_rows": 26,
        "history_step_range": [
          0,
          25
        ],
        "sha256": "d85ea77093c1a636a4bede8218f8ed050b279394788365cb92458440cf33b46d"
      }
    ],
    "evidence": "materials/multilingual-case-evidence.json",
    "observed_only": true,
    "figure": "figures/07-sao-multilingual-cases.png"
  },
  "critic_trace_analysis": {
    "source_html": "/mnt/public/storage/yuxin/harbor-verl-train/logs/harbor-swe-ohsdk-qwen3.5-35b-3nodes-veomni-swegen-1729/mix3999-sao-r3/critic_traces/index.html",
    "source_html_sha256": "7db6a38d96a4fbba875aefc0fe010bf54903bf5a6a3a98c997fe01f99a59d59a",
    "snapshot_created": "2026-09-25T06:20:32.937056+00:00",
    "prediction_stage": "before_critic_update",
    "added_at": "2026-10-07",
    "traces": 40,
    "trainer_steps": [
      112,
      131
    ],
    "index": "training_data/critic-trace-index.json",
    "arrays": "training_data/critic-trace-arrays.npz",
    "arrays_sha256": "06a9ca1e265e2df96e61a1e06c81dbfaf60e0bafcae0800e547cc802cf185f43",
    "implementation_and_config_sources": [
      {
        "path": "/mnt/public/storage/yuxin/harbor-verl-train/scripts/train/configs/fully_async_3nodes_qwen35_ohsdk_veomni_mix4k_sao_r3.env",
        "sha256": "9ceff53ac529b5ae91c2400c05062ef185de7457b14bc41da53ae38b0bf7d62d"
      },
      {
        "path": "/mnt/public/storage/yuxin/harbor-verl-train/scripts/train/configs/fully_async_3nodes_qwen35_ohsdk_veomni_swegen1729_sao.env",
        "sha256": "f196d4f22bf9787e327d704c7f316133258219e1e340df66d558b57b857c0eba"
      },
      {
        "path": "/mnt/public/storage/yuxin/harbor-verl-train/scripts/train/configs/fully_async_3nodes_qwen35_ohsdk_veomni_swegen1729.env",
        "sha256": "9a2beefa2a72b6b2eaa23d0c7f1b2928b5e663d4d9936d825f1f4aa41b42a509"
      },
      {
        "path": "/mnt/public/storage/yuxin/verl/verl/utils/critic_trace.py",
        "sha256": "5e3634cd956a1d3525b8a9d10156f230c525614e080158e8d15ca331ce59a873"
      },
      {
        "path": "/mnt/public/storage/yuxin/verl/verl/experimental/fully_async_policy/fully_async_trainer.py",
        "sha256": "1cdcb5e1be4b82e5baa03f765aa43dd2ae659daf10838e71238d0b1913be79b9"
      },
      {
        "path": "/mnt/public/storage/yuxin/verl/verl/workers/utils/padding.py",
        "sha256": "f3cc823e6ef2099075406f6fe1244498f15a775f0fac90010c0babf1c82cbe7b"
      },
      {
        "path": "/mnt/public/storage/yuxin/harbor-verl-train/scripts/tools/visualize_critic_traces.py",
        "sha256": "3623687ad9d892a5e7b5f39bfecc7dee403b2036430c25b97ff7bd97c9b6ee9d"
      }
    ],
    "display_revision": {
      "date": "2026-10-07",
      "metric": "value last minus first",
      "examples": [
        {
          "step": 127,
          "outcome": "success"
        },
        {
          "step": 118,
          "outcome": "failure"
        }
      ],
      "boundary_definition": "response-position gap and checked preceding decoded message-end marker",
      "boundary_jump_threshold": 0.1,
      "boundary_audit": "training_data/critic-turn-boundary-audit.json",
      "boundary_audit_sha256": "4cde8cefd70edba2172805c41ea6a42ca4acb7ce0ed03d22950da9b64bf82a46",
      "raw_arrays_modified": false
    }
  }
}