{
  "agent_view_tokens": 4096,
  "emoji": "🧩",
  "paper": {
    "arxiv_id": "2603.11942"
  },
  "root": {
    "children": [
      {
        "children": [],
        "file": "pages/00-judge-evidence-scorecard/page.md",
        "slug": "00-judge-evidence-scorecard",
        "title": "Judge-facing evidence scorecard"
      },
      {
        "children": [],
        "file": "pages/executive-summary/page.md",
        "slug": "executive-summary",
        "title": "Executive summary"
      },
      {
        "children": [],
        "file": "pages/claim-1/page.md",
        "slug": "claim-1",
        "title": "Claim 1: MSNN provably preserves Synthetic Nearest Neighbors' entry-wise finite-sample error bound (Theorem 4.5) and asymptotic normality with convergence rate scaling as K_MSNN^(-1/2) (Theorem 4.6) (Section 4.2)."
      },
      {
        "children": [],
        "file": "pages/claim-2/page.md",
        "slug": "claim-2",
        "title": "Claim 2: Corollary 4.10 shows the expected number of usable data subgroups under MCAR grows by a multiplicative factor of [∑_d' (p_d'/p_d)^(r+1)]^c for MSNN relative to standard SNN, formalizing the effective-sample-size gain from mixed anchor sets (Section 4.3)."
      },
      {
        "children": [],
        "file": "pages/claim-3/page.md",
        "slug": "claim-3",
        "title": "Claim 3: Corollary 4.11 shows the relative efficiency gap between data-sparse and data-rich treatment levels is reduced from quadratic to linear order under MSNN (Section 4.3)."
      },
      {
        "children": [],
        "file": "pages/claim-4/page.md",
        "slug": "claim-4",
        "title": "Claim 4: On MCAR-generated data at a low treatment probability p(d)=0.01, MSNN achieves a 4.69% feasible rate and mean relative error of 3.91e-2, versus a 0.03% feasible rate and 0.806 MRE for standard SNN (Table 1)."
      },
      {
        "children": [],
        "file": "pages/claim-5/page.md",
        "slug": "claim-5",
        "title": "Claim 5: Under MNAR data, MSNN consistently attains 3-26% feasible imputation rates versus under 5% for SNN, with 2-3x error reductions across treatment levels (Tables 2-3)."
      },
      {
        "children": [],
        "file": "pages/claim-6/page.md",
        "slug": "claim-6",
        "title": "Claim 6: Mixed anchor rows/columns are constructed via bipartite cliques spanning multiple treatment levels while preserving the target row's same-treatment data, relying on the shared latent row factor assumption (Assumption 2.5) (Section 3.2-3.3, Algorithms 2-3)."
      },
      {
        "children": [],
        "file": "pages/conclusion/page.md",
        "slug": "conclusion",
        "title": "Conclusion"
      }
    ],
    "file": "pages/index.md",
    "slug": "index",
    "title": "Reproduction: Mixed Synthetic Nearest Neighbors"
  },
  "schema_version": 1,
  "space_id": "ProCreations/repro-mixed-synthetic-nearest-neighbors-causal-matrix-completion-native",
  "tags": [
    "icml2026-repro",
    "paper-Ir6N7U5Kea"
  ],
  "title": "Reproduction: Mixed Synthetic Nearest Neighbors",
  "updated_at": "2026-07-30T10:44:00+00:00"
}