{
  "schema_version": "1.0",
  "title": "Reproduction: COPF — Deployment-Stable Counterfactual Fairness in Evolving Graphs (real TGB scale)",
  "emoji": "⚖️",
  "space_id": "snaykey/repro-rex-solvers",
  "paper": {
    "openreview_id": "2Nfr4u3pK2"
  },
  "tags": [
    "icml2026-repro",
    "paper-2Nfr4u3pK2"
  ],
  "updated_at": "2026-07-30T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: COPF — Deployment-Stable Counterfactual Fairness in Evolving Graphs (real TGB scale)",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-def33-fairness-gaps",
        "title": "COPF defines decision-counterfactual fairness gaps as disparities in benefit between exposure (shown) and non-exposure (hidden) counterfactuals across protected groups (Definition 3.3, Section 3).",
        "file": "pages/claim-1-def33-fairness-gaps/page.md",
        "children": []
      },
      {
        "slug": "claim-2-def32-gadr",
        "title": "COPF constructs graph-aware self-normalized doubly robust (GA-DR) pseudo-outcomes for estimating counterfactual quantities under temporal dependence and local interference on evolving graphs (Definition 3.2, Section 4).",
        "file": "pages/claim-2-def32-gadr/page.md",
        "children": []
      },
      {
        "slug": "claim-3-thm45-residual-oi-transfer",
        "title": "Theorem 4.5 proves that small residual outcome-indistinguishability (residual-OI) on estimated DR residuals implies bounds on the true counterfactual fairness gap, accounting for estimation and dependence error (Theorem 4.5, Section 4.2).",
        "file": "pages/claim-3-thm45-residual-oi-transfer/page.md",
        "children": []
      },
      {
        "slug": "claim-4-cor46-fairness-objective",
        "title": "Corollary 4.6 shows residual-OI control implies bounds on decision-counterfactual fairness gaps, connecting the auditing certificate to the fairness objective (Corollary 4.6, Section 4.2).",
        "file": "pages/claim-4-cor46-fairness-objective/page.md",
        "children": []
      },
      {
        "slug": "claim-5-sec6-tgb-worstcase-spikes",
        "title": "On temporal graph benchmark (TGB) streams and synthetic bipartite data, COPF reduces worst-case spikes in exposure-counterfactual group disparities during policy shifts with only modest degradation in ranking utility, as reported in the phase-wise results (Section 6, Figure 1).",
        "file": "pages/claim-5-sec6-tgb-worstcase-spikes/page.md",
        "children": []
      },
      {
        "slug": "claim-6-alg1-primal-dual-inactive",
        "title": "COPF is implemented via an online primal-dual control loop (Algorithm 1) that stays largely inactive when logged disparities are already small, only intervening when certificates flag structural bias (Section 4.3, Algorithm 1).",
        "file": "pages/claim-6-alg1-primal-dual-inactive/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  },
  "revision": 1
}