{
  "paper": {
    "arxiv_id": "2608.25635",
    "title": "DCEO"
  },
  "dataset": {
    "name": "MovieLens-1M",
    "users": 320,
    "items": 480
  },
  "setup": {
    "seed": 42,
    "full_catalog": true,
    "proxy_objectives": 4,
    "model_selection": "validation-only grid; held-out test evaluated once",
    "candidate_count": 20
  },
  "variants": {
    "fixed multi-objective fusion": {
      "hit_at_10": 0.146875,
      "ndcg_at_10": 0.07127401477576525,
      "head_share_at_10": 0.20375
    },
    "DCEO core": {
      "hit_at_10": 0.15,
      "ndcg_at_10": 0.07095207788296207,
      "head_share_at_10": 0.19875
    }
  },
  "relative": {
    "hit_at_10_percent": 2.1276595744680775,
    "ndcg_at_10_percent": -0.4516890115086558,
    "head_share_at_10_percent": -2.453987730061338
  },
  "diagnostics": {
    "proxy_outcome_correlation": -0.017584595429621274,
    "estimated_relative_causal_effect": -8.854896202385033e-07,
    "serving_critic_calls": 0,
    "selected_causal_gain": 0.75,
    "selected_temperature": 0.75,
    "selected_validation_fitness": 0.07402786918098403,
    "validation_grid_summary": {
      "primary_objective": "NDCG@10 + 0.15 * Hit@10",
      "causal_gains": [
        0.1,
        0.2,
        0.35,
        0.5,
        0.75
      ],
      "temperatures": [
        0.75,
        1.0,
        1.25,
        1.5
      ],
      "test_evaluations": 1
    }
  },
  "paper_results": {
    "online_gmv_lift_percent": 0.36,
    "ab_test_days": 41
  },
  "scope": "在 MovieLens-1M 上执行上下文相关 simplex actor、用户级 proxy 聚合与相对干预诊断；线上只运行 actor。未复刻淘宝私有搜索日志、长期 GMV critic 或生产流量。",
  "runtime": {
    "requested_device": "auto",
    "cpu_threads": null,
    "platform": "Darwin arm64"
  },
  "seed": 42,
  "schema_version": 2,
  "manifest": {
    "adapter_key": "dceo",
    "arxiv_id": "2608.25635",
    "title": "DCEO: Direct Causal Effect Optimization for Long-Term User Value Modeling in E-commerce Search",
    "paper_url": "https://arxiv.org/abs/2608.25635",
    "track": "recommendation",
    "organization": "Taobao & Tmall Group, Alibaba",
    "published": "2026-08-26",
    "code_url": null,
    "topics": [
      "search-ranking",
      "long-term-value",
      "causal-optimization",
      "multi-objective-ranking"
    ],
    "local_code_dir": "src/auto_research/reproductions/dceo",
    "fidelity": "core_mechanism",
    "evaluation_tier": "l2_public_dataset",
    "datasets": [
      "MovieLens-1M"
    ],
    "baseline": "fixed multi-objective log fusion over the same four proxy scores",
    "metrics": [
      "Hit@10",
      "NDCG@10",
      "head share@10",
      "estimated relative causal effect"
    ],
    "default_seeds": [
      42
    ],
    "budget": "adapter-defined fixed budget",
    "device_capabilities": [
      "cpu"
    ],
    "online_evidence": [
      {
        "product": "Alibaba e-commerce search",
        "metric": "GMV",
        "lift_percent": 0.36,
        "traffic": "41-day production A/B test",
        "source_url": "https://arxiv.org/html/2608.25635v1#S5.SS7",
        "source_location": "Section 5.7",
        "experiment_duration": "41 days",
        "retrieved_at": "2026-08-28"
      }
    ],
    "selection_exception": null,
    "evolve_operators": []
  },
  "provenance": {
    "created_at": "2026-08-27T16:29:37.531928+00:00",
    "code_commit": "cfb01264c6575f3d1634226940df384a3632a036",
    "python": "3.14.5",
    "platform": "macOS-26.5.2-arm64-arm-64bit-Mach-O",
    "dataset_dir": "/Users/bytedance/Documents/git_daiwk/auto-research/data",
    "dataset_fingerprint": "e66c261317a7b3179720ef3e3f5f79d7a204b135d1f217e2911b153ad2ce50a0",
    "packages": {
      "numpy": "2.4.3"
    },
    "artifact_path": "docs/reproductions/2608.25635-dceo/metrics/movielens-1m-seed42.json"
  },
  "evaluation_protocol": {
    "tier": "l2_public_dataset",
    "tier_label": "L2 公开数据集训练",
    "seeds": [
      42
    ],
    "budget": "paper-specific",
    "formal_comparison": false,
    "claim_policy": "single/few-seed smoke result; do not claim a stable improvement"
  },
  "reproduction_fidelity": {
    "level": "core_mechanism",
    "label": "核心机制复现",
    "description": "论文中心算法被实际执行，但生产模型、私有特征或基础设施未复刻。",
    "omitted_core_components": [
      "Alibaba private search logs",
      "production long-term GMV critic",
      "online serving stack"
    ]
  },
  "manifest_ref": "reproduction:dceo"
}