{ "schema_version": 1, "title": "Reproduction: Set-Preserving Calibration from Conformal P-Values to E-Values", "emoji": "📏", "space_id": "snaykey/repro-embedding-collapse", "paper": { "arxiv_id": "2606.03600", "openreview_id": "jNv4sl4YZH" }, "tags": [ "icml2026-repro", "paper-jNv4sl4YZH" ], "updated_at": "2026-07-28T06:02:04+00:00", "root": { "slug": "index", "title": "Reproduction: Set-Preserving Calibration from Conformal P-Values to E-Values", "file": "pages/index.md", "children": [ { "slug": "executive-summary", "title": "Executive summary", "file": "pages/executive-summary/page.md", "children": [] }, { "slug": "claim-1-definition-2-2-set-preserving", "title": "A calibrator is defined as set-preserving when converting conformal p-values at level alpha into e-values yields prediction sets identical to thresholding the e-values directly at 1/alpha (Section 2.1, Definition 2.2)", "file": "pages/claim-1-definition-2-2-set-preserving/page.md", "children": [] }, { "slug": "claim-2-proposition-2-3-theorem-2-6-equation-9", "title": "Among left-continuous p-to-e calibrators, only the all-or-nothing calibrator can be exactly set-preserving, which motivates a new sigmoid-based P2E construction (Proposition 2.3, Theorem 2.6, Equation 9)", "file": "pages/claim-2-proposition-2-3-theorem-2-6-equation-9/page.md", "children": [] }, { "slug": "claim-3-theorem-2-6-exact-smooth-invertible", "title": "The proposed sigmoid-based P2E calibrator is exact (expectation equal to 1), smooth, invertible, strictly positive, and strictly dominates the all-or-nothing calibrator in aggregation settings (Theorem 2.6)", "file": "pages/claim-3-theorem-2-6-exact-smooth-invertible/page.md", "children": [] }, { "slug": "claim-4-proposition-4-1-eccp-coverage", "title": "Applied to e-Cross-Conformal Prediction (ECCP), the P2E calibrator preserves exact 1-alpha coverage, whereas standard cross-conformal-prediction variants only guarantee approximate 1-2alpha coverage (Proposition 4.1)", "file": "pages/claim-4-proposition-4-1-eccp-coverage/page.md", "children": [] }, { "slug": "claim-5-proposition-4-2-weca-coverage", "title": "Applied to Weighted Conformal Aggregation (WECA), data-dependent weighted merging of e-values across multiple models retains valid coverage (Proposition 4.2)", "file": "pages/claim-5-proposition-4-2-weca-coverage/page.md", "children": [] }, { "slug": "claim-6-section-5-smaller-sets", "title": "Empirically, ECCP using the proposed P2E calibrator produces smaller prediction sets than baseline p-to-e conversion methods while maintaining valid coverage (Section 5)", "file": "pages/claim-6-section-5-smaller-sets/page.md", "children": [] }, { "slug": "conclusion", "title": "Conclusion", "file": "pages/conclusion/page.md", "children": [] } ] } }