{
  "schema_version": 2,
  "date": "2026-08-24",
  "scope": [
    "2608.18606-onemodel",
    "2608.21236-rare",
    "2608.11674-gcpo",
    "2608.21292-auso",
    "2607.26369-clockrope",
    "2607.27475-oneshot-index",
    "2607.24789-next-vlm",
    "2606.26859-agentx"
  ],
  "results": {
    "onemodel": {
      "source": "../reproductions/2608.18606-onemodel/metrics/movielens-1m-seed42.json",
      "baseline_ndcg_at_10": 0.12851715765416122,
      "method_ndcg_at_10": 0.08701532450732896,
      "relative_change": -0.322928
    },
    "rare": {
      "source": "../reproductions/2608.21236-rare/metrics/routing-mini-suite-seed42.json",
      "baseline_route_agreement": 0.82373046875,
      "method_route_agreement": 1.0,
      "method_route_flip_rate": 0.0
    },
    "gcpo": {
      "source": "../post-training/2608.11674-gcpo/metrics/arithmetic-smoke-seed42.json",
      "baseline_accuracy": 0.1953125,
      "final_accuracy": 0.6328125,
      "principal_overlap_removed": 0.995594
    },
    "auso": {
      "source": "../agent-research/2608.21292-auso/metrics/planbench-mini-seed42.json",
      "baseline_average_cost": 0.655,
      "method_average_cost": 0.5954579618,
      "relative_cost_change": -0.0909
    },
    "clockrope": {
      "source": "../reproductions/2607.26369-clockrope/metrics/movielens-100k-seed42.json",
      "baseline_ndcg_at_10": 0.0540054678,
      "method_ndcg_at_10": 0.0527646212,
      "relative_change": -0.022976313383
    },
    "oneshot-index": {
      "source": "../reproductions/2607.27475-oneshot-index/metrics/movielens-100k-seed42.json",
      "baseline_ndcg_at_10": 0.0143519359,
      "method_ndcg_at_10": 0.0132467605,
      "relative_change": -0.077005323059
    },
    "next-vlm": {
      "source": "../reproductions/2607.24789-next-vlm/metrics/movielens-1m-seed42.json",
      "baseline_ndcg_at_10": 0.0997484294,
      "method_ndcg_at_10": 0.1200258533,
      "relative_change": 0.203285646131
    },
    "agentx": {
      "source": "../agent-research/2606.26859-agentx/metrics/planbench-mini-seed42.json",
      "joint_success": 1.0,
      "average_cost": 0.587,
      "closed_loops": 12,
      "experiment_memory_reuses": 108
    }
  },
  "claim_boundary": "The aggregate is an index over per-paper source metrics, not a shared cross-task leaderboard. OneModel, ClockRoPE and OneShot retain negative public-data results; NEXT is a positive single-seed diagnostic; RARE, GCPO, AUSO and AgentX are mechanism-level mini-suite results.",
  "manifest_ref": "experiments:recent-papers-20260824-seed42",
  "evaluation_protocol": {
    "tier": "l1_mechanism",
    "seeds": [
      42
    ],
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "provenance": {
    "artifact_path": "docs/experiments/recent-papers-20260824-seed42.json",
    "dataset_fingerprint": {
      "onemodel": "MovieLens-1M:leave-two-out:seed42",
      "rare": "deterministic-moe-routing-mini-suite:seed42",
      "gcpo": "arithmetic-smoke:120-steps:seed42",
      "auso": "planbench-mini:120-episodes:seed42",
      "clockrope": "MovieLens-100K:leave-two-out:seed42",
      "oneshot-index": "MovieLens-100K:leave-two-out:seed42",
      "next-vlm": "MovieLens-1M:leave-two-out:seed42",
      "agentx": "planbench-mini:120-episodes:seed42"
    }
  }
}
