{
  "frozen_at": "2026-07-18T23:46:57.951556+00:00",
  "git_sha": "f93f4b5578c221d3d2665f9b305f3e86759be862+dirty",
  "spec": {
    "analysis": "campaign.contest.analyze",
    "hypotheses": [
      {
        "id": "H-contest",
        "prediction": {
          "ci_excludes": 0.0,
          "comparator": ">",
          "effect": null,
          "metric": "contested_raterspread_spearman",
          "threshold": 0.2
        },
        "scoreboard_row": "T7",
        "statement": "contested-direction loading predicts per-item human rater disagreement"
      },
      {
        "id": "H-contest-vs-ensemble",
        "prediction": {
          "ci_excludes": null,
          "comparator": "<",
          "effect": null,
          "metric": "ensemble_minus_contested_spearman",
          "threshold": 0.05
        },
        "scoreboard_row": null,
        "statement": "single-model contested loading comes within 0.05 of the 5x-cost ensemble baseline"
      }
    ],
    "id": "campaign-values-contest",
    "kill_criteria": [
      {
        "comparator": "<=",
        "id": "K-contest",
        "metric": "contested_raterspread_spearman",
        "threshold": 0.0
      }
    ],
    "science": "S11-values",
    "subjects": {
      "datasets": [
        "helpsteer3-pref-2000"
      ],
      "extra": {
        "slice_hashes": {
          "helpsteer3-pref-2000": "ds:2693a5ed9f17528a453bb42c436f22ea"
        }
      },
      "organisms": [],
      "signals": [
        "skywork-v2-llama31-8b",
        "ensemble-5"
      ]
    },
    "title": "Contested-direction loading predicts human rater disagreement",
    "version": 1
  },
  "spec_hash": "spec:2a1ead40948d8a58ae00bf095cad321f",
  "study_id": "study:campaign-values-contest@v1#2a1ead40"
}