[
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.0375555,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/saxpy/20261015T140009726759Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/saxpy/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0634134,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/saxpy/20261015T140009685216Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/saxpy/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0310822,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/saxpy/20261015T140009676438Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/saxpy/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.10163610000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fuzz/20261015T140036376601Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fuzz/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.07934730000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fuzz/20261015T140037158883Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fuzz/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.08042189999999999,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fuzz/20261015T140043149565Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fuzz/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.0533202,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fir/20261015T140114229680Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fir/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0551861,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fir/20261015T140122894463Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fir/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0660696,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fir/20261015T140129804185Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fir/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.0956128,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/softmax_hecbench/20261015T140149682209Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/softmax_hecbench/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0654626,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/softmax_hecbench/20261015T140155303264Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/softmax_hecbench/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.07278839999999999,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/softmax_hecbench/20261015T140209712445Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/softmax_hecbench/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.0310195,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/saxpy/20261015T140242429623Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/saxpy/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0571157,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/saxpy/20261015T140317048494Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/saxpy/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.026588,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/saxpy/20261015T140319717474Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/saxpy/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.09393710000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fuzz/20261015T140331033141Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fuzz/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.07523980000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fuzz/20261015T140346954104Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fuzz/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.045461699999999994,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fuzz/20261015T140351500406Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fuzz/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0608785,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fir/20261015T140416929213Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fir/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0467032,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fir/20261015T140425430521Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fir/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0615547,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fir/20261015T140426106831Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fir/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 7,
    "cost_usd": 0.08912769999999999,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/softmax_hecbench/20261015T140455395034Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/softmax_hecbench/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0581554,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/softmax_hecbench/20261015T140456935878Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/softmax_hecbench/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "max_iterations",
    "iterations": 10,
    "cost_usd": 0.1093652,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/softmax_hecbench/20261015T140509252051Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/softmax_hecbench/run2",
    "correct_and_clean": false,
    "claimed_done": false,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0371767,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/saxpy/20261015T140542321604Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/saxpy/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.05902,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/saxpy/20261015T140554594207Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/saxpy/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.025439299999999998,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/saxpy/20261015T140614483665Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/saxpy/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0748682,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fuzz/20261015T140630329105Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fuzz/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0780169,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fuzz/20261015T140642146687Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fuzz/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.0741203,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fuzz/20261015T140642835295Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fuzz/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0618123,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fir/20261015T140709472216Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fir/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.047808199999999995,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fir/20261015T140722282969Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fir/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0587779,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fir/20261015T140725753810Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fir/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 9,
    "cost_usd": 0.1382694,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/softmax_hecbench/20261015T140749090477Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/softmax_hecbench/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0614205,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "FAIL: 0/1 case(s) match; first failure: 332679 of 524288 element(s) outside tolerance",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/softmax_hecbench/20261015T140754104910Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/softmax_hecbench/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.0949677,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/softmax_hecbench/20261015T140803055591Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/softmax_hecbench/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.0319656,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/saxpy/20261015T140838775702Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/saxpy/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0568975,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/saxpy/20261015T140904523642Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/saxpy/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0257269,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/saxpy/20261015T140908579774Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/saxpy/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.081511,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fuzz/20261015T140913951242Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fuzz/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.07656560000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fuzz/20261015T140935323036Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fuzz/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.06985889999999999,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fuzz/20261015T140938453880Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fuzz/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0622256,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fir/20261015T140959322742Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fir/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.046218199999999994,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fir/20261015T141018783076Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fir/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0455695,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fir/20261015T141022779001Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fir/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.08683260000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/softmax_hecbench/20261015T141046807510Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.059148900000000004,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/softmax_hecbench/20261015T141050752568Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.0752057,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/softmax_hecbench/20261015T141059191554Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.030677,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/saxpy/20261015T141133658592Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/saxpy/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.059211400000000004,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/saxpy/20261015T141150356507Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/saxpy/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0268751,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/saxpy/20261015T141154815065Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/saxpy/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0758943,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fuzz/20261015T141205135231Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fuzz/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.07528180000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fuzz/20261015T141224719300Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fuzz/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.0762514,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fuzz/20261015T141228650642Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fuzz/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.059852699999999995,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/fir/20261015T141247749132Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/fir/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0464032,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/fir/20261015T141303409793Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/fir/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0591084,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/fir/20261015T141313221515Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/fir/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0670268,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/toolkit/softmax_hecbench/20261015T141329305375Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/toolkit/softmax_hecbench/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0639643,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "FAIL: 0/1 case(s) match; first failure: 332679 of 524288 element(s) outside tolerance",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline/softmax_hecbench/20261015T141335093946Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline/softmax_hecbench/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-sonnet-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.0785195,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T140008/baseline_shell/softmax_hecbench/20261015T141349681687Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-sonnet-5-5/baseline_shell/softmax_hecbench/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0028012650000000003,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/saxpy/20261015T141450041158Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/saxpy/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0031787150000000004,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/saxpy/20261015T141450168239Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/saxpy/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0025786549999999996,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/saxpy/20261015T141450166660Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/saxpy/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00240042,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fuzz/20261015T141515623946Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fuzz/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0024706949999999997,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fuzz/20261015T141527264440Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fuzz/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.004162385,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fuzz/20261015T141528567354Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fuzz/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "max_iterations",
    "iterations": 10,
    "cost_usd": 0.00661088,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fir/20261015T141550523169Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fir/run1",
    "correct_and_clean": true,
    "claimed_done": false,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.003880985,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fir/20261015T141551652218Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fir/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.005650955,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fir/20261015T141622316501Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fir/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00437382,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/softmax_hecbench/20261015T141636387523Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/softmax_hecbench/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.00530843,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/softmax_hecbench/20261015T141648627995Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/softmax_hecbench/run1",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.00626048,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/softmax_hecbench/20261015T141723487415Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/softmax_hecbench/run1",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 7,
    "cost_usd": 0.003109025,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/saxpy/20261015T141733200262Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/saxpy/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.0021053649999999997,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "candidate exited with code 1",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/saxpy/20261015T141749122398Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/saxpy/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.00289105,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/saxpy/20261015T141821925337Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/saxpy/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00211382,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fuzz/20261015T141825487715Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fuzz/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0020710700000000004,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fuzz/20261015T141843155051Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fuzz/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 10,
    "cost_usd": 0.00533573,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fuzz/20261015T141904247578Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fuzz/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0039770050000000005,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fir/20261015T141906398575Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fir/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.003398365,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fir/20261015T141916051689Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fir/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.0044757600000000005,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fir/20261015T141940216087Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fir/run2",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00360114,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/softmax_hecbench/20261015T141957637031Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/softmax_hecbench/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.005336655,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/softmax_hecbench/20261015T142009138831Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/softmax_hecbench/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.005473865,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/softmax_hecbench/20261015T142029010777Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/softmax_hecbench/run2",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.002318365,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/saxpy/20261015T142049106141Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/saxpy/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.00193293,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "candidate exited with code 1",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/saxpy/20261015T142113379696Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/saxpy/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.003842285,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/saxpy/20261015T142124810509Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/saxpy/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00227025,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fuzz/20261015T142137831751Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fuzz/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.003313195,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fuzz/20261015T142139055291Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fuzz/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.004464185000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fuzz/20261015T142214108568Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fuzz/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 7,
    "cost_usd": 0.004346445,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fir/20261015T142214129273Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fir/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0026593999999999997,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fir/20261015T142228195930Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fir/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.005199785,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fir/20261015T142256327587Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fir/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.00496483,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/softmax_hecbench/20261015T142259820513Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/softmax_hecbench/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.00496795,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/softmax_hecbench/20261015T142315650929Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/softmax_hecbench/run3",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.008364395,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 9,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/softmax_hecbench/20261015T142346961094Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/softmax_hecbench/run3",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0025201150000000003,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/saxpy/20261015T142359788435Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/saxpy/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.0018595,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "candidate exited with code 1",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/saxpy/20261015T142417352338Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/saxpy/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 7,
    "cost_usd": 0.00351743,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/saxpy/20261015T142434948247Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/saxpy/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.00236394,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fuzz/20261015T142439262961Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fuzz/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.002234505,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fuzz/20261015T142501695696Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fuzz/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.00413609,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fuzz/20261015T142514102175Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fuzz/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "max_iterations",
    "iterations": 10,
    "cost_usd": 0.007837265,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fir/20261015T142516885144Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fir/run4",
    "correct_and_clean": true,
    "claimed_done": false,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.003430405,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "candidate exited with code 1",
    "helpers": 14,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fir/20261015T142536598919Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fir/run4",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.004795790000000001,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fir/20261015T142600321578Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fir/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 4,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.006348,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/softmax_hecbench/20261015T142618647890Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.004890755,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/softmax_hecbench/20261015T142632924957Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 4,
    "stop_reason": "model_finished",
    "iterations": 10,
    "cost_usd": 0.009903875,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/softmax_hecbench/20261015T142638716057Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/softmax_hecbench/run4",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "max_iterations",
    "iterations": 10,
    "cost_usd": 0.0036418,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/saxpy/20261015T142728864969Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/saxpy/run5",
    "correct_and_clean": true,
    "claimed_done": false,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.001965905,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/saxpy/20261015T142730471129Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/saxpy/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "saxpy",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.00324985,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 4.77e-07, atol 1e-06)",
    "helpers": 11,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/saxpy/20261015T142754352582Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/saxpy/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 4,
    "cost_usd": 0.002134125,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fuzz/20261015T142811994399Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fuzz/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.00216738,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fuzz/20261015T142826277384Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fuzz/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fuzz",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.00385298,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -138.5 dBFS, threshold -96.0 dBFS)",
    "helpers": 22,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fuzz/20261015T142845408203Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fuzz/run5",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 10,
    "cost_usd": 0.0053852150000000005,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/fir/20261015T142847863133Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/fir/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.003073455,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/fir/20261015T142902324912Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/fir/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "fir",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 8,
    "cost_usd": 0.00495524,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 2/2 case(s) match (worst residual peak -135.0 dBFS, threshold -96.0 dBFS)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/fir/20261015T142929842528Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/fir/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 5,
    "stop_reason": "success",
    "iterations": 7,
    "cost_usd": 0.00671959,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/toolkit/softmax_hecbench/20261015T142932258479Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/toolkit/softmax_hecbench/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.00490907,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline/softmax_hecbench/20261015T142947208137Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline/softmax_hecbench/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  },
  {
    "model": "claude-haiku-5-5",
    "example": "softmax_hecbench",
    "arm": "baseline_shell",
    "seed": 5,
    "stop_reason": "model_finished",
    "iterations": 9,
    "cost_usd": 0.00588638,
    "compiled": true,
    "verify": "ok",
    "verify_summary": "PASS: 1/1 case(s) match (worst max abs err 9.69e-08, atol 1e-06)",
    "helpers": 0,
    "note": null,
    "run_dir": "/home/user/oneapi-agent-toolkit/runs/ablation-20261015T141448/baseline_shell/softmax_hecbench/20261015T143021256072Z",
    "evidence_dir": "docs/ablation/round2/runs/claude-haiku-5-5/baseline_shell/softmax_hecbench/run5",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false
  }
]