[
  {
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.0371807,
    "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": "runs/ablation-20261015T134507/toolkit/saxpy/20261015T134509755349Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/saxpy/run1"
  },
  {
    "example": "saxpy",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.062224,
    "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": "runs/ablation-20261015T134507/baseline/saxpy/20261015T134509678991Z",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true,
    "evidence_dir": "docs/ablation/runs/baseline/saxpy/run1"
  },
  {
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 10,
    "cost_usd": 0.1397868,
    "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": "runs/ablation-20261015T134507/toolkit/fuzz/20261015T134509816592Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fuzz/run1"
  },
  {
    "example": "fuzz",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.07526619999999999,
    "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": "runs/ablation-20261015T134507/baseline/fuzz/20261015T134509723118Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/fuzz/run1"
  },
  {
    "example": "fir",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.06872189999999999,
    "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": "runs/ablation-20261015T134507/toolkit/fir/20261015T134536446050Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fir/run1"
  },
  {
    "example": "fir",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0543586,
    "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": "runs/ablation-20261015T134507/baseline/fir/20261015T134543277249Z",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true,
    "evidence_dir": "docs/ablation/runs/baseline/fir/run1"
  },
  {
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 1,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.0775897,
    "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": "runs/ablation-20261015T134507/toolkit/softmax_hecbench/20261015T134545911476Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/softmax_hecbench/run1"
  },
  {
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 1,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0696201,
    "compiled": true,
    "verify": "fail",
    "verify_summary": "FAIL: 0/1 case(s) match; first failure: 332679 of 524288 element(s) outside tolerance",
    "helpers": 0,
    "note": null,
    "run_dir": "runs/ablation-20261015T134507/baseline/softmax_hecbench/20261015T134610868099Z",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true,
    "evidence_dir": "docs/ablation/runs/baseline/softmax_hecbench/run1"
  },
  {
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.036039,
    "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": "runs/ablation-20261015T134507/toolkit/saxpy/20261015T134612871245Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/saxpy/run2"
  },
  {
    "example": "saxpy",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 3,
    "cost_usd": 0.0209182,
    "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": "runs/ablation-20261015T134507/baseline/saxpy/20261015T134615323762Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/saxpy/run2"
  },
  {
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 7,
    "cost_usd": 0.0754081,
    "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": "runs/ablation-20261015T134507/toolkit/fuzz/20261015T134630871341Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fuzz/run2"
  },
  {
    "example": "fuzz",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.035958699999999996,
    "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": "runs/ablation-20261015T134507/baseline/fuzz/20261015T134636996571Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/fuzz/run2"
  },
  {
    "example": "fir",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.060647,
    "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": "runs/ablation-20261015T134507/toolkit/fir/20261015T134641735000Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fir/run2"
  },
  {
    "example": "fir",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0469982,
    "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": "runs/ablation-20261015T134507/baseline/fir/20261015T134653885970Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/fir/run2"
  },
  {
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 2,
    "stop_reason": "success",
    "iterations": 9,
    "cost_usd": 0.1402632,
    "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": "runs/ablation-20261015T134507/toolkit/softmax_hecbench/20261015T134702831657Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/softmax_hecbench/run2"
  },
  {
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 2,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.064232,
    "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": "runs/ablation-20261015T134507/baseline/softmax_hecbench/20261015T134710048335Z",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true,
    "evidence_dir": "docs/ablation/runs/baseline/softmax_hecbench/run2"
  },
  {
    "example": "saxpy",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 5,
    "cost_usd": 0.0315093,
    "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": "runs/ablation-20261015T134507/toolkit/saxpy/20261015T134723266103Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/saxpy/run3"
  },
  {
    "example": "saxpy",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 5,
    "cost_usd": 0.057592300000000006,
    "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": "runs/ablation-20261015T134507/baseline/saxpy/20261015T134723962318Z",
    "correct_and_clean": false,
    "claimed_done": true,
    "claimed_but_wrong": true,
    "evidence_dir": "docs/ablation/runs/baseline/saxpy/run3"
  },
  {
    "example": "fuzz",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 8,
    "cost_usd": 0.0814583,
    "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": "runs/ablation-20261015T134507/toolkit/fuzz/20261015T134751070117Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fuzz/run3"
  },
  {
    "example": "fuzz",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 6,
    "cost_usd": 0.07489589999999999,
    "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": "runs/ablation-20261015T134507/baseline/fuzz/20261015T134752363946Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/fuzz/run3"
  },
  {
    "example": "fir",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 6,
    "cost_usd": 0.060945400000000004,
    "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": "runs/ablation-20261015T134507/toolkit/fir/20261015T134757064505Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/fir/run3"
  },
  {
    "example": "fir",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.046440699999999994,
    "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": "runs/ablation-20261015T134507/baseline/fir/20261015T134817134168Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/fir/run3"
  },
  {
    "example": "softmax_hecbench",
    "arm": "toolkit",
    "seed": 3,
    "stop_reason": "success",
    "iterations": 9,
    "cost_usd": 0.089747,
    "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": "runs/ablation-20261015T134507/toolkit/softmax_hecbench/20261015T134831889611Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/toolkit/softmax_hecbench/run3"
  },
  {
    "example": "softmax_hecbench",
    "arm": "baseline",
    "seed": 3,
    "stop_reason": "model_finished",
    "iterations": 4,
    "cost_usd": 0.0590226,
    "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": "runs/ablation-20261015T134507/baseline/softmax_hecbench/20261015T134832724254Z",
    "correct_and_clean": true,
    "claimed_done": true,
    "claimed_but_wrong": false,
    "evidence_dir": "docs/ablation/runs/baseline/softmax_hecbench/run3"
  }
]