ENTRANT

A HuggingFace mirror of the ENTRANT dataset (Zenodo record 10667088, CC-BY-4.0): 6.7 million structured financial tables extracted from ~330,000 SEC EDGAR filings spanning 10 filing types.

Original paper: Gialitsis et al., Scientific Data 2024, 10.1038/s41597-024-03605-5.

License

CC-BY-4.0 (dataset annotations and structural metadata).

The underlying source documents are SEC EDGAR filings -- publicly filed corporate documents. The factual content of financial statements is not copyrightable under Feist v. Rural Telephone Service (1991); the Nature Scientific Data publication provides strong institutional backing for the CC-BY-4.0 claim on this corpus.

Cite the original ENTRANT paper if you use this dataset (see Citation below).

Configs

One HuggingFace config per SEC filing type. Each config is a single train split (the authors did not define evaluation splits; partition as needed).

Config	Filing type	Description
10-K	Annual report	Large accelerated filers, ~5.9 GB ZIP
10-Q	Quarterly report	Most common filing, ~9.4 GB ZIP
20-F	Foreign annual report	Non-US issuers, ~556 MB ZIP
S-1	IPO registration	Initial public offerings, ~179 MB ZIP
S-4	Merger/acquisition	M&A registrations, ~886 MB ZIP
8-K	Current report	Material events, ~347 MB ZIP
497	Mutual fund prospectus	Investment companies, ~29 MB ZIP
485BPOS	Fund registration	Post-effective amendments, ~19 MB ZIP
10-KT	Transition annual report	Fiscal-year change filers, ~18 MB ZIP
18-K	Foreign government annual	Sovereign / foreign government, ~37 KB ZIP

Schema

Field	Type	Description
table_id	string	{cik}_{accession_number}_{table_index}
filing_type	string	SEC filing type (e.g. 10-K)
cik	string	SEC Central Index Key
filing_year	string	Year of filing
accession_number	string	SEC accession number
title	string	Table title (may be empty)
n_rows	int32	Number of rows (from RangeAddress)
n_cols	int32	Number of columns (from RangeAddress)
n_header_rows	int32	Number of top header rows
n_header_cols	int32	Number of left header columns
html	string	Table rendered as HTML with rowspan/colspan
cells	list[struct]	Per-cell records (anchor cells only; see below)
language	string	Language of content

Cell struct fields: row, col, rowspan, colspan (int32); text (string); is_header (bool, top header); is_attribute (bool, left header). Only anchor cells (top-left corner of merged spans) are included; interior cells of a merged region are omitted.

Usage

from datasets import load_dataset

# Stream annual reports without downloading all 17 GB
ds = load_dataset(
    "rootsautomation/ENTRANT",
    "10-K",
    split="train",
    streaming=True,
)
for row in ds:
    print(row["title"], row["n_rows"], row["n_cols"])
    break

# Load a smaller config fully into memory
ds_18k = load_dataset("rootsautomation/ENTRANT", "18-K", split="train")

Citation

@article{gialitsis2024entrant,
  title   = {ENTRANT: A Large Financial Dataset for Table Understanding},
  author  = {Gialitsis, Nikolaos and Tzoumerkas, Konstantinos
             and Dalamagas, Theodore},
  journal = {Scientific Data},
  volume  = {11},
  year    = {2024},
  doi     = {10.1038/s41597-024-03605-5},
}