{
  "schema": "gemma-iterations/1",
  "source_revision": "3978c71d44a6a562d70a705c60fb2fb6dc1edf60",
  "observed": "2026-11-14 12:35:09 EST",
  "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/paper/recomputed.json",
  "protocol_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/README.md",
  "bars": {
    "ctrl_margin": 5.0,
    "s1_harmful": -6.0,
    "s1_promote": 3.0,
    "s2_min": 4.0,
    "total_harmful": -10.0,
    "total_pass": 10.0
  },
  "champion_generation": 0,
  "confirmation_note": "This export contains screening aggregates, not final-confirmation or golden evidence. Consult the protocol history for confirmation results.",
  "pending_batches": [
    {
      "batch": "b005",
      "scored": false
    },
    {
      "batch": "b010",
      "scored": false
    }
  ],
  "experiments": [
    {
      "id": "0007",
      "date": "2026-11-12",
      "author": "claude",
      "title": "Deep-thought gate (dose 0.1) plus the model s own end-of-thought neurons at 9 times their write",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0007-claude-deep-gate-b01-end-neurons-x9.md",
      "results": [
        {
          "batch": "x005",
          "arm": "e0007",
          "stage": null,
          "tasks": 120,
          "solved": 74,
          "base_solved": 75,
          "repeat_solved": 73,
          "gain": 0.0,
          "gain_pp": 0.0,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0008",
      "date": "2026-11-12",
      "author": "claude",
      "title": "Deep-thought gate: one neuron per layer 30-48 detects depth >= 800 inside a thought and writes the end-of-thought state (dose 0.3)",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0008-claude-deep-thought-gate-g3-b03.md",
      "results": [
        {
          "batch": "b002",
          "arm": "e0008",
          "stage": 1,
          "tasks": 120,
          "solved": 67,
          "base_solved": 73,
          "repeat_solved": 77,
          "gain": -8.0,
          "gain_pp": -6.667,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Harmful screen"
        }
      ],
      "batches": [
        "b002"
      ]
    },
    {
      "id": "0009",
      "date": "2026-11-12",
      "author": "claude",
      "title": "Own end-of-thought neurons x25: scale the write of the 20 late-layer MLP neurons whose write lens is the end-of-thought token (weights only)",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0009-claude-end-neurons-x25.md",
      "results": [
        {
          "batch": "x005",
          "arm": "e0009",
          "stage": null,
          "tasks": 120,
          "solved": 3,
          "base_solved": 75,
          "repeat_solved": 73,
          "gain": -71.0,
          "gain_pp": -59.167,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0010",
      "date": "2026-11-12",
      "author": "claude",
      "title": "Early deep-thought gate (fitted on depth >= 400, dose 0.1) plus own end-of-thought neurons x9: the strong dose of the gated family",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0010-claude-early-gate-b01-end-neurons-x9.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0011",
      "date": "2026-11-12",
      "author": "gpt",
      "title": "Amplify natural thought-ending writes across all layers at 0.33 percent",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0011-gpt-writer-close-amplify-033.md",
      "results": [
        {
          "batch": "b002",
          "arm": "e0011",
          "stage": 1,
          "tasks": 120,
          "solved": 65,
          "base_solved": 73,
          "repeat_solved": 77,
          "gain": -10.0,
          "gain_pp": -8.333,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Harmful screen"
        }
      ],
      "batches": [
        "b002"
      ]
    },
    {
      "id": "0012",
      "date": "2026-11-12",
      "author": "gpt",
      "title": "Amplify natural thought-ending writes across all layers at 0.50 percent",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0012-gpt-writer-close-amplify-050.md",
      "results": [
        {
          "batch": "b003",
          "arm": "e0012",
          "stage": 1,
          "tasks": 120,
          "solved": 44,
          "base_solved": 72,
          "repeat_solved": 67,
          "gain": -25.5,
          "gain_pp": -21.25,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Harmful screen"
        }
      ],
      "batches": [
        "b003"
      ]
    },
    {
      "id": "0013",
      "date": "2026-11-12",
      "author": "gpt",
      "title": "Shift common writer output toward thought endings across all layers at 0.11 percent",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0013-gpt-writer-close-shift-011.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0015",
      "date": "2026-11-12",
      "author": "gemini",
      "title": "Spectral rank-4 exact-context within-call thought-close writer subspace across all 60 layers (0.38%)",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0015-gemini-exact-close-r4-subspace.md",
      "results": [
        {
          "batch": "b003",
          "arm": "e0015",
          "stage": 1,
          "tasks": 120,
          "solved": 74,
          "base_solved": 72,
          "repeat_solved": 67,
          "gain": 4.5,
          "gain_pp": 3.75,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Advance to stage 2"
        },
        {
          "batch": "b006",
          "arm": "e0015",
          "stage": null,
          "tasks": 218,
          "solved": 119,
          "base_solved": 121,
          "repeat_solved": 110,
          "gain": 3.5,
          "gain_pp": 1.606,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b003",
        "b006"
      ]
    },
    {
      "id": "0016",
      "date": "2026-11-12",
      "author": "gemini",
      "title": "Subtractive rank-2 deep-rumination attractor damping across all 280 mid-to-late modules (0.36%)",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0016-gemini-rumination-damp-r2-net.md",
      "results": [
        {
          "batch": "x002",
          "arm": "e0016",
          "stage": null,
          "tasks": 120,
          "solved": 73,
          "base_solved": 73,
          "repeat_solved": 72,
          "gain": 0.5,
          "gain_pp": 0.417,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0017",
      "date": "2026-11-12",
      "author": "gemini",
      "title": "Coordinate-whitened GeGLU late pre-edit commit gate across 13 high-signal layers (dose 1.20)",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0017-gemini-late-preedit-commit-gate.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0018",
      "date": "2026-11-12",
      "author": "gemini",
      "title": "Dual-cap rescue stack: rank-4 exact-close writer subspace (0.38%) plus 13-layer late pre-edit commit gate",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0018-gemini-dual-budget-rescue-r5.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0020",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Deep-thought gate in all 60 layers: one neuron per layer detects depth >= 800 in a thought and writes the end-of-thought state (per-layer do",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0020-claude-net-deep-gate-close-b05.md",
      "results": [
        {
          "batch": "b004",
          "arm": "e0020",
          "stage": 1,
          "tasks": 120,
          "solved": 58,
          "base_solved": 70,
          "repeat_solved": 73,
          "gain": -13.5,
          "gain_pp": -11.25,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Harmful screen"
        }
      ],
      "batches": [
        "b004"
      ]
    },
    {
      "id": "0021",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Deep-thought gate in all 60 layers: one neuron per layer detects depth >= 800 in a thought and writes the end-of-thought state (per-layer do",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0021-claude-net-deep-gate-close-b12.md",
      "results": [
        {
          "batch": "b004",
          "arm": "e0021",
          "stage": 1,
          "tasks": 120,
          "solved": 60,
          "base_solved": 70,
          "repeat_solved": 73,
          "gain": -11.5,
          "gain_pp": -9.583,
          "corrected": true,
          "provisional": false,
          "clock": "450-second session limit",
          "task_set": "Stage 1 final list v2",
          "decision": "Harmful screen"
        }
      ],
      "batches": [
        "b004"
      ]
    },
    {
      "id": "0022",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Pre-edit commit gate in all 60 layers: deep in a thought with no edit made yet, write the first-edit thought state only (per-layer dose 0.12",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0022-claude-net-preedit-commit-b12.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0023",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Pre-edit rumination gate in all 60 layers: deep in a thought with no edit made yet, write end-of-thought plus first-edit state (per-layer do",
      "status": "parked",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0023-claude-net-preedit-commit-close-b05.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0025",
      "date": "2026-11-13",
      "author": "gpt",
      "title": "Damp the code-versus-prose style contrast in writers in all layers at 0.33 percent",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0025-gpt-code-writer-damp-r4-033.md",
      "results": [
        {
          "batch": "b007",
          "arm": "e0025",
          "stage": null,
          "tasks": 120,
          "solved": 68,
          "base_solved": 65,
          "repeat_solved": 73,
          "gain": -1.0,
          "gain_pp": -0.833,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b007"
      ]
    },
    {
      "id": "0026",
      "date": "2026-11-13",
      "author": "gpt",
      "title": "Damp the code-versus-prose style contrast in writers in all layers at 0.11 percent",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0026-gpt-code-writer-damp-r4-011.md",
      "results": [
        {
          "batch": "b007",
          "arm": "e0026",
          "stage": null,
          "tasks": 120,
          "solved": 69,
          "base_solved": 65,
          "repeat_solved": 73,
          "gain": 0.0,
          "gain_pp": 0.0,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b007"
      ]
    },
    {
      "id": "0027",
      "date": "2026-11-13",
      "author": "gpt",
      "title": "Damp the code-versus-prose style contrast in readers in all layers at 0.33 percent",
      "status": "parked",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0027-gpt-code-reader-damp-r4-033.md",
      "results": [],
      "batches": [
        "b012"
      ]
    },
    {
      "id": "0028",
      "date": "2026-11-13",
      "author": "gemini",
      "title": "Protocol-nullspace rank-4 exact thought-close writer subspace across L10-59 (0.48%)",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0028-gemini-nullspace-close-r4-midlate.md",
      "results": [
        {
          "batch": "x002",
          "arm": "e0028",
          "stage": null,
          "tasks": 120,
          "solved": 62,
          "base_solved": 73,
          "repeat_solved": 72,
          "gain": -10.5,
          "gain_pp": -8.75,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0031",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Instruction vector, think briefly: an always-on neuron per layer writes the stream shift of a 'keep your thinking short' paragraph (dose 2)",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0031-claude-icv-brief-b20.md",
      "results": [
        {
          "batch": "b008",
          "arm": "e0031",
          "stage": null,
          "tasks": 218,
          "solved": 122,
          "base_solved": 113,
          "repeat_solved": 112,
          "gain": 9.5,
          "gain_pp": 4.358,
          "corrected": true,
          "provisional": false,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        },
        {
          "batch": "x001",
          "arm": "e0031",
          "stage": null,
          "tasks": 120,
          "solved": 77,
          "base_solved": 72,
          "repeat_solved": 64,
          "gain": 9.0,
          "gain_pp": 7.5,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b008"
      ]
    },
    {
      "id": "0032",
      "date": "2026-11-13",
      "author": "claude",
      "title": "Instruction vector, think briefly: an always-on neuron per layer writes the stream shift of a 'keep your thinking short' paragraph (dose 4)",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0032-claude-icv-brief-b40.md",
      "results": [
        {
          "batch": "b009",
          "arm": "e0032",
          "stage": null,
          "tasks": 218,
          "solved": 123,
          "base_solved": 127,
          "repeat_solved": 114,
          "gain": 2.5,
          "gain_pp": 1.147,
          "corrected": true,
          "provisional": false,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        },
        {
          "batch": "x001",
          "arm": "e0032",
          "stage": null,
          "tasks": 120,
          "solved": 71,
          "base_solved": 72,
          "repeat_solved": 64,
          "gain": 3.0,
          "gain_pp": 2.5,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b009"
      ]
    },
    {
      "id": "0035",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Source-grounding instruction contrast across all 60 layers at dose 3",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0035-gpt-source-grounding-icv-b3.md",
      "results": [
        {
          "batch": "x003",
          "arm": "e0035",
          "stage": null,
          "tasks": 120,
          "solved": 79,
          "base_solved": 81,
          "repeat_solved": 82,
          "gain": -2.5,
          "gain_pp": -2.083,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0036",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Source-grounding instruction contrast across all 60 layers at dose 1",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0036-gpt-source-grounding-icv-b1.md",
      "results": [
        {
          "batch": "x003",
          "arm": "e0036",
          "stage": null,
          "tasks": 120,
          "solved": 78,
          "base_solved": 81,
          "repeat_solved": 82,
          "gain": -3.5,
          "gain_pp": -2.917,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0038",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Think-briefly instruction vector, self-consistent compile (dose 2): each layer writes only what the paragraph's mean shift still lacks",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0038-claude-fcv-brief-b20.md",
      "results": [
        {
          "batch": "x004",
          "arm": "e0038",
          "stage": null,
          "tasks": 120,
          "solved": 75,
          "base_solved": 78,
          "repeat_solved": 76,
          "gain": -2.0,
          "gain_pp": -1.667,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0039",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Instruction pair vector: think briefly + be exact about what the issue asks, compiled self-consistently into one neuron per layer (dose 2)",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0039-claude-fcv-brief-exact-b20.md",
      "results": [
        {
          "batch": "x004",
          "arm": "e0039",
          "stage": null,
          "tasks": 120,
          "solved": 79,
          "base_solved": 78,
          "repeat_solved": 76,
          "gain": 2.0,
          "gain_pp": 1.667,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0043",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Thought-gated reliability-shrunk Problem Statement spec-audit steering across all 60 MLP layers",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0043-gemini-spec-audit-thought-gate.md",
      "results": [
        {
          "batch": "x006",
          "arm": "e0043",
          "stage": null,
          "tasks": 120,
          "solved": 79,
          "base_solved": 78,
          "repeat_solved": 75,
          "gain": 2.5,
          "gain_pp": 2.083,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": []
    },
    {
      "id": "0044",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Stateful two-neuron post-edit spec-audit boost with thought and edit-history gating over 180 MLP matrices",
      "status": "queued_s2",
      "verdict": "not confirmed",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0044-gemini-dual-audit-postedit-boost.md",
      "results": [
        {
          "batch": "x006",
          "arm": "e0044",
          "stage": null,
          "tasks": 120,
          "solved": 82,
          "base_solved": 78,
          "repeat_solved": 75,
          "gain": 5.5,
          "gain_pp": 4.583,
          "corrected": false,
          "provisional": true,
          "clock": "Not classified by this exporter",
          "task_set": "Consult batch source",
          "decision": "Unclassified aggregate — consult source"
        }
      ],
      "batches": [
        "b011"
      ]
    },
    {
      "id": "0047",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Paired outcome writer map across all 60 layers at 1 percent",
      "status": "running_s1",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0047-gpt-paired-outcome-writers-r4-100.md",
      "results": [],
      "batches": [
        "b010"
      ]
    },
    {
      "id": "0048",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Paired outcome writer map across all 60 layers at 0.33 percent",
      "status": "running_s1",
      "verdict": "not confirmed",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0048-gpt-paired-outcome-writers-r4-033.md",
      "results": [],
      "batches": [
        "b010"
      ]
    },
    {
      "id": "0049",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Think-briefly + be-exact prompt compiled where a prompt enters: many-key rank-8 down_proj maps in 7 global-attention layers, dose 2",
      "status": "running_s1",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0049-claude-mk-pair-global-b20.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0050",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Instruction pair vector, 'be exact about what the issue asks' part written three times in layers 6 to 24, think-briefly part at dose 2",
      "status": "running_s1",
      "verdict": "not confirmed",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0050-claude-fcv-brief-exact3-b20.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0055",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "OOF-shrunk rank-8 multi-key phase teachers across 108 RoPE-free o_proj and down_proj matrices",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0055-gemini-sc-multikey-spec-writers-r8.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0056",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Capacity-regularized three-phase GeGLU controller for concise thoughts, post-edit spec audit, and pre-submit veto",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0056-gemini-tri-phase-concise-audit-veto.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0061",
      "date": "2026-11-14",
      "author": "claude",
      "title": "'Wait' contrast removed from every writer: no block output in any layer can prefer 'Wait' to the other words that open a sentence",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0061-claude-wait-contrast-full.md",
      "results": [],
      "batches": [
        "b013"
      ]
    },
    {
      "id": "0062",
      "date": "2026-11-14",
      "author": "claude",
      "title": "'Wait' contrast at one third: every writer in every layer keeps two thirds of its preference for 'Wait' over other sentence starters",
      "status": "done",
      "verdict": "null",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0062-claude-wait-contrast-third.md",
      "results": [],
      "batches": [
        "b013"
      ]
    },
    {
      "id": "0063",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Damp the code-versus-prose style contrast in readers across all 60 layers at 0.66 percent",
      "status": "done",
      "verdict": "harmful",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0063-gpt-code-reader-damp-r4-066.md",
      "results": [],
      "batches": [
        "b012"
      ]
    },
    {
      "id": "0064",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Stateful two-neuron post-edit spec-audit boost with thought and edit-history gating over 180 MLP matrices - re-entered on champion generation 1 (0031)",
      "status": "running",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0064-gemini-dual-audit-postedit-boost-g1.md",
      "results": [],
      "batches": [
        "b014"
      ]
    },
    {
      "id": "0065",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Instruction pair vector, 'be exact about what the issue asks' part written three times in layers 6 to 24, think-briefly part at dose 2 - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0065-claude-fcv-brief-exact3-b20-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0066",
      "date": "2026-11-14",
      "author": "gpt",
      "title": "Paired outcome writer map across all 60 layers at 0.33 percent - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0066-gpt-paired-outcome-writers-r4-033-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0067",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Instruction vector, think briefly: an always-on neuron per layer writes the stream shift of a 'keep your thinking short' paragraph (dose 4) - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0067-claude-icv-brief-b40-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0068",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Thought-gated reliability-shrunk Problem Statement spec-audit steering across all 60 MLP layers - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0068-gemini-spec-audit-thought-gate-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0069",
      "date": "2026-11-14",
      "author": "claude",
      "title": "Instruction pair vector: think briefly + be exact about what the issue asks, compiled self-consistently into one neuron per layer (dose 2) - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0069-claude-fcv-brief-exact-b20-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0070",
      "date": "2026-11-14",
      "author": "gemini",
      "title": "Spectral rank-4 exact-context within-call thought-close writer subspace across all 60 layers (0.38%) - re-entered on champion generation 1 (0031)",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0070-gemini-exact-close-r4-subspace-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0074",
      "date": "2026-11-14",
      "author": "grok",
      "title": "One-third tool_act plus on a fresh neuron, added on top of 0031",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0074-grok-toolact-plus-007-g1.md",
      "results": [],
      "batches": []
    },
    {
      "id": "0075",
      "date": "2026-11-14",
      "author": "grok",
      "title": "Always-on MLP bake of the tool_act plus direction at 2% of mean stream norm",
      "status": "queued",
      "verdict": "-",
      "source_url": "https://github.com/Autoresearch-Arena/gemma-4-developer-agent/blob/3978c71d44a6a562d70a705c60fb2fb6dc1edf60/research/wsp31/ledger/0075-grok-toolact-plus-020.md",
      "results": [],
      "batches": []
    }
  ]
}