{
  "description": "Provenance for HABIT-Bench website numerical content. No raw user histories or private annotations are included.",
  "writing_repository": {
    "commit": "f52bc1643c2b95c57d72588df50cb4f319c8c280",
    "files": {
      "iclr2027_conference.tex": "39529ddb3c356ac509fbc1f6740cf7070d09e299c294015654389fb967505d83",
      "sec/3_habit_bench.tex": "bfe5f060356562d225afcd94e108862a0524ee6c7be160b093168d31db62ef43",
      "sec/4_experiments.tex": "fc44ac143f54ea8d8a3ca5c4728f9440a0b3dd911aaa58d5644d166af0264c37",
      "tables/dataset_overview.tex": "e3275505d1f26829890b75f1b38754677ec28084893cd1b02256aff5c6f7dab7",
      "tables/method_provenance.tex": "cf6a76b3484d64b1962d815e83693a8018370fb23b94d3d1d8169bc88b8d98b6",
      "paper_audit/dataset_manifest.csv": "bd7fb090a26fecbfe5443725d37c5b4e8ee2bc39d622f3e79a5e51cfc21d8006"
    }
  },
  "reference_repository": {
    "commit": "242a1bf61181acd853fd281ac13810ec0b52a9a4",
    "result_document": "docs/main_experiment_results_current.md",
    "sha256": "c96ca6aed0739c8b6fbf12944502ac6f20f0631d1c71856d8a254641c4bd8c2e",
    "heading": "## Qwen3-235B-A22B 主实验 Acc（四域 × 17 项，含诊断对照）",
    "source_lines": [
      {
        "method": "No-Memory",
        "line": 237
      },
      {
        "method": "Full-Memory",
        "line": 238
      },
      {
        "method": "Recency-5",
        "line": 239
      },
      {
        "method": "Recency-10",
        "line": 240
      },
      {
        "method": "BM25-RAG",
        "line": 241
      },
      {
        "method": "Dense-RAG",
        "line": 242
      },
      {
        "method": "Temporal Hybrid-RAG",
        "line": 243
      },
      {
        "method": "Mem0",
        "line": 244
      },
      {
        "method": "A-MEM",
        "line": 245
      },
      {
        "method": "MemOS",
        "line": 246
      },
      {
        "method": "MemRL",
        "line": 247
      },
      {
        "method": "LightMem",
        "line": 248
      },
      {
        "method": "Letta",
        "line": 249
      },
      {
        "method": "MIRIX",
        "line": 250
      },
      {
        "method": "SeCom",
        "line": 251
      }
    ]
  },
  "datasets": [
    {
      "domain": "food",
      "package": "domain/food/food_habit_lifelines_final_check",
      "lifecycle": "active_candidate",
      "domain_filter": null,
      "public_sha256": {
        "lifelines": "d9feee413db8d5f2621683697f3d161be98b614c4316e3914ceb5c28206d6fa7",
        "probes": "f96348b5546f8c622cec32eb9bca62011a9562d8b89668d15b6d93bafa284f25"
      }
    },
    {
      "domain": "travel",
      "package": "domain/travel/release_candidate_v16_postrepair_repaired_r4",
      "lifecycle": "active_candidate",
      "domain_filter": null,
      "public_sha256": {
        "lifelines": "a36b27465a280804f4103b2ad28266bc6f27d4a935754e5aed1b9b5f82d1352b",
        "probes": "a89d872554bfaa35b15c683e71a30bb1ebc45ea81177e737c6935a44987a4651"
      }
    },
    {
      "domain": "finance",
      "package": "domain/finance-software/habit_bench_multidogo_finance_software_release_gated_v1_4",
      "lifecycle": "active_candidate",
      "domain_filter": "finance",
      "public_sha256": {
        "lifelines": "0c38a8a5eadc258dcf2a728e122fa9fddb943b533da4ef5be56c64926f14c320",
        "probes": "0e2f38e2cfa40bd429d06f334331b8c8050bf8e3ce6d3ffe94875e40b2e2398e"
      }
    },
    {
      "domain": "software",
      "package": "domain/finance-software/habit_bench_multidogo_finance_software_release_gated_v1_4",
      "lifecycle": "active_candidate",
      "domain_filter": "software",
      "public_sha256": {
        "lifelines": "0c38a8a5eadc258dcf2a728e122fa9fddb943b533da4ef5be56c64926f14c320",
        "probes": "0e2f38e2cfa40bd429d06f334331b8c8050bf8e3ce6d3ffe94875e40b2e2398e"
      }
    }
  ],
  "dataset_statistics": {
    "status": "manifest-verified; independently recounted during website audit",
    "manifest": "paper_audit/dataset_manifest.csv",
    "table": "tables/dataset_overview.tex:8-13"
  },
  "experiment_results": {
    "status": "document-reported",
    "metric": "Per-domain exact-choice accuracy (%) and unweighted four-domain macro accuracy (%)",
    "versions": "Food final; Finance v1.4; Software v1.4; Travel v16",
    "displayed": "All eight memory methods, five session-retrieval baselines, and two evaluator controls from this comparison. Private diagnostic oracles are outside this display.",
    "limitations": [
      "Not independently recomputed for this website.",
      "Method-specific recorded context budgets.",
      "Official-code adapters and official-source adaptations are not upstream-native end-to-end evaluations.",
      "25% is a theoretical uniform-choice reference, not a significance threshold.",
      "No matched transfer-gap claim is established here."
    ]
  },
  "illustration": {
    "status": "Newly authored conceptual example; not an evaluation item or private gold annotation"
  },
  "release": {
    "status": "Public project website with arXiv paper, Hugging Face dataset, and public GitHub code repository.",
    "paper": "https://arxiv.org/pdf/2610.24520",
    "verified_paper_version": "2610.24520v2",
    "code": "https://github.com/LUMIA-Group/habit-bench",
    "figure_sources": "figure-provenance.json",
    "dataset": "https://huggingface.co/datasets/hollyming/HABIT-Bench"
  },
  "publication_crosscheck": {
    "paper_version": "2610.24520v2",
    "source_url": "https://arxiv.org/src/2610.24520v2",
    "table": "tables/dual_model_results.tex",
    "table_sha256": "de899085110b06a82e4a5a906f5a07c6ff23897ede8631359f473db3fe6e45d9",
    "scope": "All 15 displayed non-oracle Qwen3-235B-A22B rows, including four-domain macro averages, matched against the published table. The interactive chart displays the Qwen subset, not the GLM comparison.",
    "status": "document-reported; cross-checked against published table, not raw-recomputed"
  }
}