The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
index_settings: struct<docstore_compression: string, docstore_blocksize: int64>
child 0, docstore_compression: string
child 1, docstore_blocksize: int64
segments: list<item: struct<segment_id: string, max_doc: int64, deletes: null>>
child 0, item: struct<segment_id: string, max_doc: int64, deletes: null>
child 0, segment_id: string
child 1, max_doc: int64
child 2, deletes: null
schema: list<item: struct<name: string, type: string, options: struct<indexed: bool, fieldnorms: bool, fast: (... 112 chars omitted)
child 0, item: struct<name: string, type: string, options: struct<indexed: bool, fieldnorms: bool, fast: bool, stor (... 100 chars omitted)
child 0, name: string
child 1, type: string
child 2, options: struct<indexed: bool, fieldnorms: bool, fast: bool, stored: bool, indexing: struct<record: string, f (... 55 chars omitted)
child 0, indexed: bool
child 1, fieldnorms: bool
child 2, fast: bool
child 3, stored: bool
child 4, indexing: struct<record: string, fieldnorms: bool, tokenizer: string>
child 0, record: string
child 1, fieldnorms: bool
child 2, tokenizer: string
child 5, precision: string
opstamp: int64
tantivy: string
max_doc_chars: int64
inputs: list<item: string>
child 0, item: string
characters: int64
chunk_chars: int64
public: bool
by_source: struct<packet:fav2: int64, packet:bigfinancebench: int64, packet:diligencebench: int64, corpus:edgar (... 28 chars omitted)
child 0, packet:fav2: int64
child 1, packet:bigfinancebench: int64
child 2, packet:diligencebench: int64
child 3, corpus:edgar: int64
child 4, corpus:wiki: int64
documents: int64
normalizer_version: string
built_at: timestamp[s]
seconds: int64
to
{'built_at': Value('timestamp[s]'), 'documents': Value('int64'), 'characters': Value('int64'), 'chunk_chars': Value('int64'), 'max_doc_chars': Value('int64'), 'public': Value('bool'), 'by_source': {'packet:fav2': Value('int64'), 'packet:bigfinancebench': Value('int64'), 'packet:diligencebench': Value('int64'), 'corpus:edgar': Value('int64'), 'corpus:wiki': Value('int64')}, 'inputs': List(Value('string')), 'normalizer_version': Value('string'), 'tantivy': Value('string'), 'seconds': Value('int64')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
index_settings: struct<docstore_compression: string, docstore_blocksize: int64>
child 0, docstore_compression: string
child 1, docstore_blocksize: int64
segments: list<item: struct<segment_id: string, max_doc: int64, deletes: null>>
child 0, item: struct<segment_id: string, max_doc: int64, deletes: null>
child 0, segment_id: string
child 1, max_doc: int64
child 2, deletes: null
schema: list<item: struct<name: string, type: string, options: struct<indexed: bool, fieldnorms: bool, fast: (... 112 chars omitted)
child 0, item: struct<name: string, type: string, options: struct<indexed: bool, fieldnorms: bool, fast: bool, stor (... 100 chars omitted)
child 0, name: string
child 1, type: string
child 2, options: struct<indexed: bool, fieldnorms: bool, fast: bool, stored: bool, indexing: struct<record: string, f (... 55 chars omitted)
child 0, indexed: bool
child 1, fieldnorms: bool
child 2, fast: bool
child 3, stored: bool
child 4, indexing: struct<record: string, fieldnorms: bool, tokenizer: string>
child 0, record: string
child 1, fieldnorms: bool
child 2, tokenizer: string
child 5, precision: string
opstamp: int64
tantivy: string
max_doc_chars: int64
inputs: list<item: string>
child 0, item: string
characters: int64
chunk_chars: int64
public: bool
by_source: struct<packet:fav2: int64, packet:bigfinancebench: int64, packet:diligencebench: int64, corpus:edgar (... 28 chars omitted)
child 0, packet:fav2: int64
child 1, packet:bigfinancebench: int64
child 2, packet:diligencebench: int64
child 3, corpus:edgar: int64
child 4, corpus:wiki: int64
documents: int64
normalizer_version: string
built_at: timestamp[s]
seconds: int64
to
{'built_at': Value('timestamp[s]'), 'documents': Value('int64'), 'characters': Value('int64'), 'chunk_chars': Value('int64'), 'max_doc_chars': Value('int64'), 'public': Value('bool'), 'by_source': {'packet:fav2': Value('int64'), 'packet:bigfinancebench': Value('int64'), 'packet:diligencebench': Value('int64'), 'corpus:edgar': Value('int64'), 'corpus:wiki': Value('int64')}, 'inputs': List(Value('string')), 'normalizer_version': Value('string'), 'tantivy': Value('string'), 'seconds': Value('int64')}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ARB finance simulated internet
A frozen copy of the web pages that the finance benchmarks of Applied RSI Bench (ARB) draw on
(fav2, which is Vals AI Finance Agent v2, BigFinanceBench and DiligenceBench), with a search
index. Pages keep their real URLs. ARB's finance capsules download it and give the evaluated
model two tools over it, web_search and fetch_url, which follow the request and response
formats of Tavily's /search and /extract.
Each benchmark sees the web as of its own date (fav2 2026-03-01, BigFinanceBench 2026-05-28, DiligenceBench 2026-05-21): pages dated after it are hidden from search and from reading, and SEC filing indexes and XBRL company facts lose their later rows. 12,062 of the 249,613 pages carry no date (Wikipedia articles and many agency pages, fetched in October 2026); they are always visible and can hold later facts.
| pages | 249,613 from 63 websites |
| text | 27.5 billion characters |
| dated pages | 237,551, from 2001-02-09 to 2026-10-02 |
| files | docs.sqlite (18.4 GB), index/ (tantivy 0.26.2, 8.3 GB), MANIFEST.json |
| built | 2026-10-05 by ARB's simweb build --public |
What it holds
| source | pages | billion characters |
|---|---|---|
| SEC EDGAR filings of S&P 1500 companies since 2019 | 209,376 | 14.88 |
| ARB data packet: DiligenceBench (redistributable build) | 14,672 | 4.71 |
| ARB data packet: fav2 (redistributable build) | 13,839 | 5.42 |
| ARB data packet: BigFinanceBench (redistributable build) | 10,486 | 2.42 |
| Wikipedia articles of S&P 1500 companies | 1,240 | 0.04 |
SEC pages by form: 8-K 166,622, 10-Q 34,858, 10-K 13,155, DEF 14A 11,869, 8-K/A 2,340, 6-K 2,262, DEFM14A 1,832, 4 1,145. Pages over 20 MB (dataset dumps such as Medicare data files) are left out of the web; the data packets keep them.
Only sources whose terms allow redistribution are included, the same rule as the data packets: SEC filings, works of the U.S. federal government, statutes and court opinions, and EU and Wikipedia texts under their licenses. Share-price histories are not among them, so questions that need market prices cannot be fully answered from this web.
Contamination check
The web holds no benchmark questions: a 16-gram check (ARB's python -m contamination overlap rule) finds 0 of the 227 questions with 25% or more of their 16-word spans in it
(fav2 0 of 27, BigFinanceBench 0 of 50, DiligenceBench
0 of 150). 3 questions share a shorter quote with a filing (at most 5.1% of
a question). The benchmarks' repositories and datasets were never fetched.
Use
hf download junlinw/arb-finance-web --repo-type dataset --local-dir finance-web
from simweb.web import SimWeb # applied-rsi-bench: simweb/ (capsules: finweb/web.py)
web = SimWeb("finance-web", as_of="2026-05-28")
web.search("Snowflake fiscal 2026 product revenue", max_results=5, ticker="SNOW")
web.extract(["https://www.sec.gov/..."], query="product revenue")
ARB pins the exact commit its capsules download; pass --revision <commit> to get the same
bytes.
Layout
MANIFEST.json inputs, counts by source, build settings
docs.sqlite docs(id, url, host, title, date, tickers, cik, form, company, source, nchars,
nchunks) and chunks(doc, no, text): each page's text in ~2,000-character
zlib-compressed slices
index/ tantivy index of the chunks, with url, host, date, tickers and form fields
Sources and licenses
| license (from the builder's table) | pages | billion characters |
|---|---|---|
| public-record (sec.gov) | 238,507 | 24.22 |
| us-government-work | 9,642 | 3.19 |
| CC-BY-SA-4.0 (Wikipedia) | 1,248 | 0.04 |
| eu-reuse-with-attribution | 170 | 0.03 |
| government-edict | 46 | < 0.01 |
SEC filings are public records. Works of the U.S. federal government are not under copyright
in the United States. Wikipedia texts are CC BY-SA 4.0, converted to plain text: credit
Wikipedia's contributors and share alike; each page's URL is in docs.sqlite. EU texts follow
the Commission's reuse policy (Decision 2011/833/EU), which asks for attribution. These are not
official versions; check the publisher before relying on a text. To have a page removed, open
a discussion on this dataset.
- Downloads last month
- -