{
  "schema_version": 1,
  "date": "2026-10-18",
  "distribution_version": "1.1.1",
  "status": "data_handoff_complete_rescoring_pending",
  "summary": "Original Opus data recovered; new DeepSeek paraphrase and baseline runs completed and verified in private collaborator datasets. Sentence-segmentation correction and rescoring are pending.",
  "accepted_snapshot": {
    "date": "2026-10-01",
    "unchanged": true,
    "openreview_url": "https://openreview.net/forum?id=1E20ig92Zi",
    "pdf_sha256": "ee5301d4067b76d143aef274d5cf8339bb60c4aa1fbf0f36a63575cea083c8ad",
    "scores_status": "unchanged",
    "findings_describe": "October 1 accepted snapshot; not the October 18 reruns"
  },
  "rescoring": {
    "status": "pending",
    "owner": "Vittoria Lanzo",
    "scope": "Sentence boundaries after curly closing quotes; re-score affected model data and the new DeepSeek runs before updating findings."
  },
  "archive": {
    "status": "unchanged",
    "path": "data/released/hf-datasets.zip",
    "sha256": "4e1e7aced0203b9c20e4d52f04cc1111be9a9de8ffab58a42d551250bc83f47e",
    "bytes": 17979599,
    "raw_model_families": 8,
    "opus_included": false,
    "deepseek_october_runs_included": false,
    "qwen_rows": 2987,
    "public_package_contains_private_raw_data": false
  },
  "deepseek": {
    "api_model_alias": "deepseek-v4-pro",
    "response_model_alias": "deepseek-v4-pro",
    "documented_snapshot": "DeepSeek-V4-Pro-0813",
    "snapshot_date": "2026-08-13",
    "snapshot_note": "The provider documented this later snapshot at rerun time; responses returned the alias, not a pinned July model identifier.",
    "model_documentation_url": "https://api-docs.deepseek.com/quick_start/pricing/",
    "settings": {
      "extra_body": {
        "reasoning_effort": "max",
        "thinking": {
          "type": "enabled"
        }
      },
      "max_tokens": 32000,
      "model": "deepseek-v4-pro",
      "stream": true,
      "stream_options": {
        "include_usage": true
      }
    },
    "input": {
      "filename": "prompts_200_human.jsonl",
      "historical_revision_was_unpinned": true,
      "repo_id": "burstiness-engine/paper-dataset",
      "revision": "3f27ce645ce4f21aa91bd9b8b52a9ca33214fdee",
      "sha256": "fa73bffe8b2a312ae0902a90a3b3e72b07de74ddbc8a3dd49134c39377859ed6"
    },
    "source_git_commit": "959853c6aace188975fb907ee6946bcc8032afe2",
    "historical_prompt_reference_commit": "780641d0b09c9fea1698b649c9652fc9ec1f1b07",
    "baseline_historical_prompt_comparisons": 4200,
    "baseline_historical_prompt_mismatches": 0,
    "iteration_note": "Iterations label repeated requests. No API seed parameter was supplied.",
    "original_july_paraphrases_recovered": false,
    "token_cap_note": "The October runs use max_tokens=32000, a disclosed correction from the checked-in 16000-token configuration. The effective cap of the original July paraphrase run could not be established.",
    "comparability_note": "These are new October generations from a later model snapshot. They do not recover or exactly replicate the July paraphrase experiment."
  },
  "datasets": [
    {
      "id": "opus-original",
      "title": "Recovered original Claude Opus 4.8",
      "classification": "recovered_original",
      "rows": 3000,
      "coverage": {
        "scenarios": 200,
        "modes": [
          "control",
          "instruction",
          "fewshot",
          "plan",
          "self_refine"
        ],
        "iterations": [
          0,
          1,
          2
        ],
        "self_refine_drafts": 600,
        "self_refine_critiques": 600
      },
      "repo": "burstiness-engine/opus48-news-gen",
      "commit": "5f9e1bcdb993171dbd586f3c4b8ad1f053d3c0c9",
      "commit_url": "https://huggingface.co/datasets/burstiness-engine/opus48-news-gen/commit/5f9e1bcdb993171dbd586f3c4b8ad1f053d3c0c9",
      "url": "https://huggingface.co/datasets/burstiness-engine/opus48-news-gen/tree/5f9e1bcdb993171dbd586f3c4b8ad1f053d3c0c9/data",
      "prefix": "data",
      "access": "private_collaborator_access_required",
      "private": true,
      "files": [
        {
          "path": "data/opus-4.8.jsonl",
          "bytes": 6683873,
          "sha256": "354a2ca63409b825f525bb374e26ab91072fb38004c349085c05eb757dcf439a",
          "download_verified": true
        },
        {
          "path": "data/opus-4.8.batchstate.json",
          "bytes": 3184473,
          "sha256": "1322ae3ea182f759c541afffd8ae7d828d9251e8bdb14dc170a549499fe89096",
          "download_verified": true
        }
      ],
      "settings": {
        "reasoning_effort": "high",
        "max_tokens": 32000
      },
      "original_run_date": "2026-07-13",
      "new_generation": false
    },
    {
      "id": "deepseek-paraphrase-oct18",
      "title": "New DeepSeek paraphrase run",
      "classification": "new_generation",
      "rows": 600,
      "coverage": {
        "scenarios_per_variant": 200,
        "variants": [
          "v1",
          "v2",
          "v3"
        ],
        "iterations": [
          0
        ]
      },
      "repo": "burstiness-engine/deepseek-v4-news-gen",
      "commit": "69dc8a31412b45bf06848b3b1f0ec53cfc37af34",
      "commit_url": "https://huggingface.co/datasets/burstiness-engine/deepseek-v4-news-gen/commit/69dc8a31412b45bf06848b3b1f0ec53cfc37af34",
      "url": "https://huggingface.co/datasets/burstiness-engine/deepseek-v4-news-gen/tree/69dc8a31412b45bf06848b3b1f0ec53cfc37af34/paraphrase/new-run-2026-10-18",
      "prefix": "paraphrase/new-run-2026-10-18",
      "access": "private_collaborator_access_required",
      "private": true,
      "files": [
        {
          "path": "paraphrase/new-run-2026-10-18/README.md",
          "bytes": 1992,
          "sha256": "9fe72c1330557e29e546b817810ffd7faf2455b4a91a5b97e27d618cc0c083ea",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/appendix-note.txt",
          "bytes": 1682,
          "sha256": "a439cc5c3dbd983d78558c37a69cc1b6435b18eafbf1cc20f538690a2729e5d8",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/manifest.json",
          "bytes": 4892,
          "sha256": "1462deafcfe3c5ecfffa173c1718a7f4c2b05045eb4b1ba3140fb85721fa9675",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/run-summary.json",
          "bytes": 787,
          "sha256": "534e025e2062936465a1c668835f910f703696ffef15be3afacbc1000b24b3f5",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/jobs.jsonl",
          "bytes": 413612,
          "sha256": "e6e44a49312b30968c0a1f8346a1eb66af73943f5a33921f5df83ff73158acef",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/attempts.jsonl",
          "bytes": 605575,
          "sha256": "8e18141d515d1f9727f918233948c69bd451977f7f725a273c00367e022e64a2",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/failed_cells.jsonl",
          "bytes": 5305,
          "sha256": "1ebde13d06d77bfb6a44a08809d03c3382753aa9707cef29ceab53cbfd567b2d",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/continuation-manifest.json",
          "bytes": 21692,
          "sha256": "a8f3387485fbae701b8f643598aca81d76e9b3d62e09810cb3f486c11ccc6f3e",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/repair-manifest.json",
          "bytes": 1802,
          "sha256": "27929497b603f277d61b3ca095501607a21c98282ae50780f24779052c5e33eb",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/attempt-selection.json",
          "bytes": 204619,
          "sha256": "48986f11a049726e547aee6857f1f00f0ffe19077f5c4617a36a3702728ffe8a",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/variant_v1.jsonl",
          "bytes": 4017391,
          "sha256": "c6d7521df21d991026e0ba2abc92e2eb0a62d73b89cc927bb29d27ca6992dc89",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/variant_v2.jsonl",
          "bytes": 4281309,
          "sha256": "5dbe33653a3d6f0e40cb112901712229a678e398d7f8124069972962ab655d7d",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/variant_v3.jsonl",
          "bytes": 4186643,
          "sha256": "e19df19bb46a4747179cfe9800ac46fb6deaf9a3cfdb621a244c69af1f5f9df6",
          "download_verified": true
        },
        {
          "path": "paraphrase/new-run-2026-10-18/SHA256SUMS",
          "bytes": 1084,
          "sha256": "2dd583b814d2c7f6516a158b2f6bc07f5cece8c24e4cf8a16743820567afcb73",
          "download_verified": true
        }
      ],
      "new_generation": true,
      "recovered_original": false,
      "manifest_sha256": "1462deafcfe3c5ecfffa173c1718a7f4c2b05045eb4b1ba3140fb85721fa9675",
      "run_started_at": "2026-10-18T18:34:03.938032+00:00",
      "run_completed_at": "2026-10-18T18:53:17.295909+00:00",
      "source_git_commit": "959853c6aace188975fb907ee6946bcc8032afe2",
      "validation": {
        "selected_rows": 600,
        "total_attempts": 602,
        "historical_failed_attempts": 2,
        "successful_repairs": 2,
        "unresolved_cells": 0,
        "selected_finish_reason": "stop",
        "empty_selected_outputs": 0
      }
    },
    {
      "id": "deepseek-baseline-oct18",
      "title": "New DeepSeek baseline run",
      "classification": "new_generation",
      "rows": 3000,
      "coverage": {
        "scenarios": 200,
        "modes": [
          "control",
          "instruction",
          "fewshot",
          "plan",
          "self_refine"
        ],
        "iterations": [
          0,
          1,
          2
        ],
        "selected_stages": 4200,
        "self_refine_drafts": 600,
        "self_refine_critiques": 600
      },
      "repo": "burstiness-engine/deepseek-v4-news-gen",
      "commit": "e35f0b872b489151fc2e430b879499b20092884c",
      "commit_url": "https://huggingface.co/datasets/burstiness-engine/deepseek-v4-news-gen/commit/e35f0b872b489151fc2e430b879499b20092884c",
      "url": "https://huggingface.co/datasets/burstiness-engine/deepseek-v4-news-gen/tree/e35f0b872b489151fc2e430b879499b20092884c/benchmark/new-run-2026-10-18",
      "prefix": "benchmark/new-run-2026-10-18",
      "access": "private_collaborator_access_required",
      "private": true,
      "files": [
        {
          "path": "benchmark/new-run-2026-10-18/README.md",
          "bytes": 3087,
          "sha256": "54a03d4fd1fed531ebbbed886d4e7d97baf62e74dea4574b29977ac664f229de",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/appendix-note.txt",
          "bytes": 737,
          "sha256": "e565580e76ad9b37be0cee1dd1d2b6d84a0812fe8be7ba7ff1295c75e0769ef9",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/manifest.json",
          "bytes": 6803,
          "sha256": "dbe021beca5b366d6daef8168f8d656929adf1cbdefda094e8d1c4564fca49f5",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/run-summary.json",
          "bytes": 2062,
          "sha256": "16c4436b68c6851d1dd06899fb83accf6902e06ef81074ef6fbb8f1483d08c06",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/cells.jsonl",
          "bytes": 1269660,
          "sha256": "2ee43e01b0b028e24ed67d7075ed9016e60120662aa7570ef0c12bbca032b368",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/deepseek-v4.jsonl",
          "bytes": 59145972,
          "sha256": "32fe26ada9c8c27a653d056c3102090a0aa5cee5b57916fdafc2c086a72a3145",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/selfrefine-batchstate.json",
          "bytes": 48149454,
          "sha256": "1f9871c8bc17a61d20b6f616ff4a160825ecf83b9000b275b03f2919645e892b",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/stage-requests.jsonl",
          "bytes": 14373332,
          "sha256": "95ef6e1514712adb5c28260c6b09d22a6bda2d8d48d3f12b69dbaac52fdc686b",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/attempts.jsonl",
          "bytes": 86868923,
          "sha256": "3563f8c382300ae97c4ddef885600d2b6484c254ecb9f9673548911d69803af8",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/failed-attempts.jsonl",
          "bytes": 889164,
          "sha256": "5c119edf42669a4213f175011d09c0a8f47911397486db07137315f23b1cdcb2",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/attempt-selection.json",
          "bytes": 958463,
          "sha256": "9ccdd7c34e2f6ea0cf19e1bb2b92ff4d4a141377e866cf9d1c0c652cc5948741",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/repair-manifest.json",
          "bytes": 38839,
          "sha256": "17f40d0b63c867d55401657cdf01fd4890b05542e67ceb74ed2b1ad3ac45deaa",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/repair2-manifest.json",
          "bytes": 29426,
          "sha256": "876b0a472549dd1daf7fb4b94b3a7005deb9beccbc233d6aab5ebedd1382da35",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/repair1-attempt-selection.json",
          "bytes": 957035,
          "sha256": "33adce5450fd1083feae8b3d7f701881d6c573529bca5f045a55c84078b58d05",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/repair2-attempt-selection.json",
          "bytes": 958463,
          "sha256": "9ccdd7c34e2f6ea0cf19e1bb2b92ff4d4a141377e866cf9d1c0c652cc5948741",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/transport-continuation-manifest.json",
          "bytes": 82283,
          "sha256": "57ec024ceaf3bcd5b15cb8346902c32fed0ab1e260dbb8bb56b470d0e80c0cab",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/transport-continuation-receipt.json",
          "bytes": 25009,
          "sha256": "31a081160bcec4e253acd0448c175893fe7cfb7450a59de800d6a54b8b48a025",
          "download_verified": true
        },
        {
          "path": "benchmark/new-run-2026-10-18/SHA256SUMS",
          "bytes": 1497,
          "sha256": "69648bcabf5dc21618ddc51327212e85a78dba20e3ab8a1a593636af4c0dc951",
          "download_verified": true
        }
      ],
      "new_generation": true,
      "recovered_original": false,
      "manifest_sha256": "dbe021beca5b366d6daef8168f8d656929adf1cbdefda094e8d1c4564fca49f5",
      "run_started_at": "2026-10-18T18:53:35.689907+00:00",
      "run_completed_at": "2026-10-18T19:44:10.511661+00:00",
      "source_git_commit": "959853c6aace188975fb907ee6946bcc8032afe2",
      "validation": {
        "selected_rows": 3000,
        "selected_stages": 4200,
        "total_attempts": 4271,
        "historical_failed_attempts": 71,
        "retry_calls": 71,
        "first_calls_to_downstream_stages_during_repairs": 36,
        "unresolved_cells": 0,
        "selected_finish_reason": "stop",
        "empty_selected_outputs": 0
      }
    }
  ],
  "quality": {
    "rescored": false,
    "retained_flags": [
      {
        "dataset_id": "deepseek-baseline-oct18",
        "eval_prompt_id": "self_refine",
        "scenario_id": 93,
        "iteration": 1,
        "stage": "rewrite",
        "classification": "off_topic_nonarticle_response",
        "observed_word_count": 51,
        "disposition": "Retained unchanged; no retry or exclusion.",
        "cause": "Undetermined; the recorded request matches the frozen prompt."
      }
    ],
    "attempt_history": "Original attempts, failed attempts, repairs and selected-output provenance are preserved in the private run packages."
  }
}