{
  "bench_version": 1,
  "track": "tangent_evasion_dynamics",
  "model": "rnn_small",
  "metric_direction": "lower is better",
  "n_seeds": 8,
  "baseline": {
    "best_cfg": {
      "lr": 0.01
    },
    "sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.2184290774166584
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.20245468616485596
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.14241471700370312
      }
    ],
    "full": {
      "mean": 0.1416995758190751,
      "std": 0.033042414878946644,
      "per_seed": [
        0.17624007165431976,
        0.18836471438407898,
        0.0758553221821785,
        0.12919875979423523,
        0.13717937469482422,
        0.14442023634910583,
        0.12063135206699371,
        0.16170677542686462
      ],
      "n": 8
    }
  },
  "idea": {
    "mean": 0.14497848600149155,
    "std": 0.025671386028026995,
    "per_seed": [
      0.17028073966503143,
      0.15589244663715363,
      0.09865665435791016,
      0.1305036097764969,
      0.12954431772232056,
      0.18738092482089996,
      0.15148448944091797,
      0.13608470559120178
    ],
    "n": 8
  },
  "comparison": {
    "delta_mean": 0.003278910182416439,
    "idea_wins": 4,
    "n_pairs": 8,
    "per_seed_diffs": [
      -0.00595933198928833,
      -0.032472267746925354,
      0.02280133217573166,
      0.0013048499822616577,
      -0.007635056972503662,
      0.04296068847179413,
      0.030853137373924255,
      -0.025622069835662842
    ],
    "p_value": 0.74865,
    "mde": 0.022431633117906365,
    "mde_rel_pct": 15.83041656140702,
    "verdict": "no significant win",
    "system_worked": false
  },
  "mechanism_signature": {
    "predicted": "projection keeps one-step disk penetration at numerical zero",
    "observed_baseline_max_penetration": 0.05145234614610672,
    "observed_idea_max_penetration": 0.0,
    "confirmed": true,
    "custom_track": {
      "name": "tangent_evasion_dynamics",
      "file": "bench_tangent_stage2.py",
      "domain": "dynamics"
    },
    "idea_sweep": [
      {
        "cfg": {
          "lr": 0.001
        },
        "mean": 0.22645806334912777
      },
      {
        "cfg": {
          "lr": 0.003
        },
        "mean": 0.2063459139317274
      },
      {
        "cfg": {
          "lr": 0.01
        },
        "mean": 0.14497848600149155
      }
    ]
  }
}