{
  "schema_version": "1.0",
  "title": "Reproduction: Robust Sequential Experimental Design for A/B Testing",
  "emoji": "🔬",
  "space_id": "snaykey/repro-robust-seq-ab-testing",
  "paper": {
    "arxiv_id": null,
    "openreview_id": "Q7lEZjtDKO"
  },
  "tags": [
    "icml2026-repro",
    "paper-Q7lEZjtDKO"
  ],
  "updated_at": "2026-07-29T00:00:00+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: Robust Sequential Experimental Design for A/B Testing",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1",
        "title": "Theorem 1 formulates the optimal sequential treatment allocation problem in contextual bandits as a dynamic program with Markov state (Delta_i, Gamma_i) representing covariate imbalance statistics (Theorem 1, Section 3).",
        "file": "pages/claim-1/page.md",
        "children": []
      },
      {
        "slug": "claim-2",
        "title": "Theorem 2 and Theorem 3 extend the dynamic-programming formulation to dynamic settings with carryover effects, decomposing the problem into an across-day Bellman recursion over N days and a within-day finite-horizon MDP solvable via reinforcement learning (Theorem 2, Theorem 3, Section 4).",
        "file": "pages/claim-2/page.md",
        "children": []
      },
      {
        "slug": "claim-3",
        "title": "The proposed robust sequential design (RSD) bounds the worst-case mean squared error of the estimated treatment effect under model misspecification via an orthogonalization technique, remaining valid when the true outcome model contains unknown nonlinear components (Section 3).",
        "file": "pages/claim-3/page.md",
        "children": []
      },
      {
        "slug": "claim-4",
        "title": "In synthetic contextual bandit experiments with both additive and interactive treatment effects, and in dynamic settings with time horizons T in {6, 12} using 400-1000 simulation replications, RSD consistently outperforms random assignment, switchback designs, and Neyman-balanced procedures (Section 5).",
        "file": "pages/claim-4/page.md",
        "children": []
      },
      {
        "slug": "claim-5",
        "title": "RSD is evaluated on real-world data from a ride-sharing company in addition to synthetic experiments, demonstrating effectiveness beyond simulation (Section 5).",
        "file": "pages/claim-5/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  }
}