logbook.json

{
  "schema_version": 1,
  "title": "Reproduction: Questioning the Coverage-Length Metric in Conformal Prediction",
  "emoji": "📐",
  "space_id": "snaykey/repro-conformal-coverage-length",
  "paper": {
    "openreview_id": "r3h23Jv26a"
  },
  "tags": [
    "icml2026-repro",
    "paper-r3h23Jv26a"
  ],
  "updated_at": "2026-08-01T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: Questioning the Coverage-Length Metric in Conformal Prediction",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "claim-1-prejudicial-trick-theorem-6",
        "title": "The Prejudicial Trick (Algorithm 1) constructs prediction intervals by returning a null (empty or minimal) interval with probability (1-p) and an adjusted conformal interval with probability p, provably preserving valid marginal coverage per Theorem 6 while shrinking average interval length (Section on Prejudicial Trick, Algorithm 1, Theorem 6)",
        "file": "pages/claim-1-prejudicial-trick-theorem-6/page.md",
        "children": []
      },
      {
        "slug": "claim-2-synthetic-pt-vcp-table-1",
        "title": "On synthetic data, PT applied to Vanilla Conformal Prediction (PT-VCP) reduces average interval length from 22.894 to 22.614 at alpha=0.10 while preserving nominal coverage (Table 1)",
        "file": "pages/claim-2-synthetic-pt-vcp-table-1/page.md",
        "children": []
      },
      {
        "slug": "claim-3-real-benchmarks-table-2",
        "title": "On real-world regression benchmarks (including MEPS and BIO), PT-VCP reduces interval length in 9 of 10 datasets while maintaining 90% marginal coverage (Table 2)",
        "file": "pages/claim-3-real-benchmarks-table-2/page.md",
        "children": []
      },
      {
        "slug": "claim-4-interval-stability-proposition-2",
        "title": "Despite valid marginal coverage, the Prejudicial Trick yields highly unstable predictions: Proposition 2 proves its interval stability metric IS(C_PT) = p(1-p)(E(L))^2 > 0, meaning the same input can receive completely different interval outputs across repeated conformal calibration runs (Definition 1, Proposition 2)",
        "file": "pages/claim-4-interval-stability-proposition-2/page.md",
        "children": []
      },
      {
        "slug": "claim-5-localized-cp-proposition-1",
        "title": "The proposed Interval Stability metric can flag methods, such as localized conformal prediction, that implicitly exploit PT-like randomness to shrink length, since Proposition 1 shows PT is a special case of localized CP when local scale estimators output extreme values (Definition 1, Proposition 1)",
        "file": "pages/claim-5-localized-cp-proposition-1/page.md",
        "children": []
      },
      {
        "slug": "executive-summary",
        "title": "executive-summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  }
}