Datasets:

delip
/

sep11docs

pretty_name: NYC September 11 Document Release
license: other
license_name: nyc-public-records
license_details: >-
  Public records released by the City of New York through its September 11
  document release portal (https://sept11documents.cityofnewyork.us/). No
  additional license is asserted by the dataset maintainers.
language:
  - en
task_categories:
  - document-question-answering
  - text-retrieval
tags:
  - september-11
  - 9/11
  - nyc
  - government-records
  - pdf
  - archives
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: doc_id
      dtype: string
    - name: pdf_url
      dtype: string
    - name: location_tag
      dtype: string
    - name: pdf_path
      dtype: string
    - name: pdf_content
      dtype: pdf
  splits:
    - name: train
      num_examples: 24441
      num_bytes: 36604656024
  download_size: 36607511743
  dataset_size: 36604656024

NYC September 11 Document Release

This dataset is a snapshot of the PDF documents published by the City of New York on its September 11 document release portal:

https://sept11documents.cityofnewyork.us/

It contains 24441 PDF files (36.6 GB) exactly as they were served by the portal, together with the URL each file was fetched from and the archival location (agency, box, folder) the portal lists for it.

What is in the data

Column	Type	Description
`doc_id`	string	Document identifier, the PDF filename without the `.pdf` extension (for example `NYC-WTC_000000001`). Unique per row.
`pdf_url`	string	The portal URL the PDF was downloaded from.
`location_tag`	string	Archival location as shown on the portal, with levels separated by `›` (for example `Environmental Protection, Dept. of › DEP Box 04 › 524`).
`pdf_path`	string	Relative path of the file in the original download tree (`pdfs/<bucket>/<doc_id>.pdf`). Kept for provenance and for joining with local copies.
`pdf_content`	pdf	The PDF file itself, stored as bytes. Rendered with thumbnails in the Dataset Viewer and decoded to a `pdfplumber` object by the `datasets` library.

There is a single train split. The data is stored as 36 Parquet shards under data/.

Where it came from and when

The documents were made public by the City of New York under the administration of Mayor Zohran Mamdani through the September 11 document release portal. The City built and maintains the portal; this dataset simply mirrors what it published.

All files were downloaded from https://sept11documents.cityofnewyork.us/ on 2026-09-09. The list of document URLs and their location tags was collected from the portal's public listing on the same day.

This is a snapshot

The City of New York may add, replace, or remove documents on the portal after this initial tranche was collected. This dataset reflects only what was available on 2026-09-09 and will not track later changes automatically.

The intent of publishing it here is to make large scale processing, search, and study of the collection easier than fetching tens of thousands of individual files from the portal. If a specific document matters to your work, please verify it against the live copy on the portal, which is the authoritative source.

Disclaimer

Our interest is in making these records widely accessible given their national importance. We have not processed, filtered, redacted, OCRed, or otherwise altered the documents in any way. Each file is byte for byte what the portal served at download time. Any text, metadata, or redactions inside the PDFs come from the City's release, not from us.

The records were released by the City of New York. If you notice personally identifiable information (PII) or other content that you believe should not be public, please contact the City through the document release portal at https://sept11documents.cityofnewyork.us/ rather than the dataset maintainers, since only the City can correct the source release.

Loading

from datasets import load_dataset

ds = load_dataset("delip/sep11docs", split="train")
row = ds[0]
print(row["doc_id"], row["location_tag"])
pdf = row["pdf_content"]            # pdfplumber.pdf.PDF
print(pdf.pages[0].extract_text())

To skip decoding and work with raw bytes:

from datasets import Pdf
ds = ds.cast_column("pdf_content", Pdf(decode=False))
raw = ds[0]["pdf_content"]["bytes"]

To stream without downloading everything:

ds = load_dataset("delip/sep11docs", split="train", streaming=True)
for row in ds.take(5):
    print(row["doc_id"])

Acknowledgements

We thank the City of New York (@nycgov) and the Office of the Mayor of New York City (@NYCMayor) for making these records publicly available through the September 11 document release portal.

Citation

If you use this dataset, please cite it as:

@misc{rao2026sep11docs,
  author       = {Rao, Delip},
  title        = {NYC September 11 Document Release: A Snapshot of the City of New York's September 11 Document Portal},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/delip/sep11docs}},
  note         = {Snapshot of https://sept11documents.cityofnewyork.us/ taken on 2026-09-09}
}