{
  "id": "opaque-rotating-test-sets",
  "type": "entity",
  "name": "Opaque rotating test sets",
  "summary": "Held-out evaluation tasks that the evaluated system (and its training pipeline) never sees, refreshed on a schedule. Opacity prevents direct optimization against the benchmark; rotation limits contamination and slow overfitting as public data leaks into training corpora. The price is comparability: rotated scores are trends, not fixed leaderboards.",
  "locale": "en",
  "tags": [
    "evaluation",
    "benchmarks",
    "contamination",
    "goodhart"
  ],
  "relations": [
    {
      "rel": "defends_against",
      "target": "goodhart-resistance"
    },
    {
      "rel": "related",
      "target": "behavioral-canaries"
    }
  ],
  "questions": [
    "Why keep an eval set private?",
    "How do I stop my agent from overfitting its own benchmark?"
  ],
  "claims": [
    {
      "id": "c1",
      "text": "OpenAI's evals framework explicitly supports private evals built on your own data without public exposure, so teams can grade models without publishing the test set.",
      "sources": [
        "openai-evals"
      ],
      "confidence": 0.9
    },
    {
      "id": "c2",
      "text": "Optimizing against a known, static metric degrades what the metric measures — Goodhart dynamics with at least four distinct mechanisms — which is the failure opaque, rotating sets are designed to resist.",
      "sources": [
        "goodhart-variants"
      ],
      "confidence": 0.85
    },
    {
      "id": "c3",
      "text": "METR publishes datasets and research on evaluation integrity threats for autonomous AI systems, treating benchmark integrity as a first-class research problem.",
      "sources": [
        "metr"
      ],
      "confidence": 0.8
    }
  ],
  "takeaways": [],
  "faqs": [],
  "evidence_tier": "secondary",
  "evidence": {
    "level": "industry_observation",
    "source_types": [
      "industry_observation",
      "paper"
    ]
  },
  "moat_flag": false,
  "winning_edge": "Connects benchmark privacy and rotation to the specific Goodhart mechanisms they defeat, and states the comparability cost honestly — most treatments advocate private sets without pricing the trade.",
  "confidence": 0.8,
  "last_verified": "2026-08-08",
  "canonical_url": "https://agentreliability.dev/k/opaque-rotating-test-sets",
  "api_url": "https://agentreliability.dev/api/k/opaque-rotating-test-sets.json",
  "jsonld": {
    "@context": "https://schema.org",
    "name": "Opaque rotating test sets",
    "description": "Held-out evaluation tasks that the evaluated system (and its training pipeline) never sees, refreshed on a schedule. Opacity prevents direct optimization against the benchmark; rotation limits contamination and slow overfitting as public data leaks into training corpora. The price is comparability: rotated scores are trends, not fixed leaderboards.",
    "url": "https://agentreliability.dev/k/opaque-rotating-test-sets",
    "license": "https://spdx.org/licenses/CC-BY-4.0.html",
    "dateModified": "2026-08-08",
    "citation": [
      {
        "@type": "CreativeWork",
        "name": "openai/evals — framework for evaluating LLMs and LLM systems",
        "url": "https://github.com/openai/evals"
      },
      {
        "@type": "CreativeWork",
        "name": "Categorizing Variants of Goodhart's Law",
        "url": "https://arxiv.org/abs/1803.04585"
      },
      {
        "@type": "CreativeWork",
        "name": "METR — Model Evaluation & Threat Research",
        "url": "https://metr.org/"
      }
    ],
    "author": {
      "@type": "Person",
      "name": "Santiago Santa María Morales",
      "jobTitle": "practitioner — harness engineering and agent evaluation in production"
    },
    "@type": "Article",
    "headline": "Opaque rotating test sets"
  }
}