{
  "schema_version": "1.1",
  "id": "archive:https://arxiv.org/abs/2611.14514v1",
  "slug": "2611-14514v1-1f7jn00",
  "url": "https://feed7.dev/p/2611-14514v1-1f7jn00",
  "title": "Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations",
  "why_included": "Contrastive Discovery asks research agents to compare competing explanations, so rejecting one hypothesis can still identify a plausible alternative using the same evidence.",
  "summary": "The method has agents propose mutually exclusive, exhaustive explanations and score experiments by how evidence shifts belief across them. Across **six domains**, contrast sets resolved **73%** of candidates that pointwise agents had rejected; guiding search resolved **67% more questions** under the same experiment budget.",
  "practical_implication": "For builders of research agents, make the alternatives explicit when a test can falsify a hypothesis without identifying what is true. Existing agents can apply contrast sets to previously collected evidence as well as use them to guide new experiments.",
  "agent_context": "The method has agents propose mutually exclusive, exhaustive explanations and score experiments by how evidence shifts belief across them. Across **six domains**, contrast sets resolved **73%** of candidates that pointwise agents had rejected; guiding search resolved **67% more questions** under the same experiment budget.\n\nFor builders of research agents, make the alternatives explicit when a test can falsify a hypothesis without identifying what is true. Existing agents can apply contrast sets to previously collected evidence as well as use them to guide new experiments.\n\nIn a blinded study, **38 experts** selected the same explanation as the agent on 78% of resolved questions. The material describes scientific discovery tasks; it does not show that this approach improves general coding-agent performance.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2611.14514v1",
    "published_at": "2026-11-10T18:58:43.000Z"
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "agent",
  "domains": [
    "research"
  ],
  "topics": [
    "reasoning",
    "agent-reliability"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "In a blinded study, **38 experts** selected the same explanation as the agent on 78% of resolved questions. The material describes scientific discovery tasks; it does not show that this approach improves general coding-agent performance."
  ],
  "connected_context": {
    "meaning": "Contrastive Discovery makes uncertainty among explanations explicit: experiments are selected for how they distinguish alternatives, which can recover candidates that pointwise testing rejects. Its reported benefits concern scientific discovery tasks and do not establish gains for general coding agents.",
    "corpus_size": 1059,
    "generated_at": "2026-11-11T11:09:28.413Z",
    "connections": [
      {
        "title": "CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.22511v1",
        "feed7_url": "https://feed7.dev/p/2607-22511v1-0mgsdh3",
        "reason": "Both target scientific research agents, but CausalForge checks derivations and audits claim semantics, whereas Contrastive Discovery uses competing explanations to choose informative experiments; the methods address distinct parts of research reliability."
      },
      {
        "title": "AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification",
        "source_name": "arXiv",
        "source_url": "https://arxiv.org/abs/2607.11849v1",
        "feed7_url": "https://feed7.dev/p/2607-11849v1-1sjdl4n",
        "reason": "AdvancedMathBench tests proof generation and rejection of invalid proofs, while Contrastive Discovery tests whether evidence separates explanations; together they distinguish formal verification from experimental discrimination in research-agent evaluation."
      }
    ]
  },
  "lifecycle": "Current",
  "published_at": "2026-11-10T18:58:43.000Z",
  "modified_at": "2026-11-10T18:58:43.000Z",
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2611-14514v1-1f7jn00",
    "json": "https://feed7.dev/p/2611-14514v1-1f7jn00.json",
    "markdown": "https://feed7.dev/p/2611-14514v1-1f7jn00.md"
  }
}