The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 84, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
FrontierFinance raw corpus
Study material for the frontierfinance task of Applied RSI Bench (ARB): the documents an
analyst would read to answer the investor research queries of
FrontierFinance (SEC filings and
their exhibits, among them earnings releases and presentations filed on Form 8-K, foreign
issuers' 20-F and 6-K, Federal Reserve, BLS, BEA, Treasury and other U.S. federal pages and
data), as plain text where possible. An ARB post-training agent gets it read-only at
/home/agent/workspace/raw_corpus/frontierfinance/ before its model is evaluated.
This is a redistributable build. It keeps only documents from sources whose terms allow
redistribution (src/raw_corpus/hosts.toml): SEC filings, works of the U.S. federal
government, EU and Wikipedia texts under their licenses, and a separate layer of pages
made from public Hugging Face datasets that state an open licence (below). The first build
fetched 23,219 documents and left out the 3,661 from other sources (earnings-call transcript
sites, company investor-relations sites, news and data vendors). Share prices and earnings-call
transcripts are now partly here (SEC price data and the open-licence layer); consensus
estimates, sell-side reports and news are not, so rubrics that need them cannot be met from the
corpus alone.
The corpus holds no FrontierFinance queries or rubrics: the builder refuses every URL of the
benchmark's repository, dataset, paper and site. A 16-gram check (ARB's python -m contamination overlap) agrees: 0 of the 126 test queries long enough to check have 25 % or
more of their 16-word spans in it. The 8 pages the finance web sim-ff4 withheld after its own
16-gram check against the rubrics (7 of them were among this corpus's sources) are left out too.
| file | docs | unpacked → archive | tree_sha256 |
|---|---|---|---|
L.tar.gz |
180,816 | 9.56 GB → 2.16 GB | fca912c4108a7ce1946829aa7781161476acdaeb16c6a13e1234bd36e129a8b7 |
What changed from the first build (d22ab41a)
The first build's 19,558 documents are all here; 59 of them, U.S. government tables, are the
re-dated versions below. Added on 2026-10-27, so that the corpus holds
what ARB's finance web holds of FrontierFinance's sources (the web's revisions sim-ff3 and
sim-ff4, leonmodal/arb-finance-web-ff); each group is its own map entry in PACKET.json:
| entry | files | what |
|---|---|---|
ff-sec-companies |
12,634 | SEC filings of the companies the queries name, in the 550 days up to each query date (EX-99 exhibits, 10-K, 10-Q, 20-F, 40-F, 6-K, proxies, registration statements, Forms 4 near each query date) |
ff-sec-history |
20,135 | SEC filings of the same companies since 2019 (earnings, Reg FD and other-event 8-Ks with exhibits, periodic reports) |
ff-etf-holdings |
29 | N-PORT holdings reports of the XBI and XOP ETFs |
wiki-current-events |
1,216 | Wikipedia's daily Current events pages, 2023-01-01 to 2026-04-30 (CC BY-SA 4.0) |
ff4-sec-companies |
13,371 | SEC filings of the companies the other three ARB finance tasks name, 2017 on |
ff4-xbrl-reports |
77,797 | SEC XBRL financial data, one page per 10-K, 10-Q, 20-F or 40-F |
ff4-ftd-prices |
12,504 | daily closing prices from SEC's fails-to-deliver data, one page per security and year |
ff4-sec-price-points |
5,391 | share prices in SEC filings: insiders' open-market trades (Forms 4) by month, issuer repurchase prices |
ff4-gov-series |
154 | U.S. government tables (Treasury yields, Federal Reserve H.15, EIA prices) re-dated by row; they replace the first build's copies of the same URLs |
open-licence-defeatbeta |
18,059 | earnings-call transcripts and daily share prices, open-licence layer |
open-licence-kurry |
27 | earnings-call transcripts, open-licence layer |
The open-licence dataset layer
18,086 files are made from public Hugging Face datasets that state an open licence, each
pinned to one revision, under the dataset's own URLs
(huggingface.co/datasets/<id>/blob/<revision>/<file>?<row key>). It is separate: its files
are the entries open-licence-*, and each file's meta in PACKET.json has layer
(open-licence-dataset), source_dataset, revision and licence. The licence is as
stated by the dataset, and the original source as stated by its card:
| dataset | revision | licence as stated | original source as stated by the card |
|---|---|---|---|
defeatbeta/yahoo-finance-data |
86776ce5 |
ODC-BY | Yahoo! Finance, Nasdaq and the U.S. Department of the Treasury |
kurry/sp500_earnings_transcripts |
f3ded372 |
MIT | not stated |
No news item is in the corpus (the datasets' news files are left out).
How it was built (the first build)
ARB's raw_corpus builder (map → scout → script → list → build), 2026-10-24:
- map: one agent (Claude Fable 5.1) read the benchmark and listed what its work needs: 9 domains, 106 entries.
- scout: one agent (Claude Sonnet 5.5) per entry found where its documents live online; 78 entries were scouted, and 28 whose sources are all private publishers (transcript sites, price and screener sites, news, ratings, sell-side and industry reports, company websites, accounting standards) were not.
- script: list scripts for 4 new websites; the others were already in the shared host registry.
- list: code kept up to 1,000 documents per entry: 23,245 URLs.
- build: code fetched each URL once (robots.txt honored, per-site rate limits), turned HTML and PDF into text, and hashed every file; 14 URLs were refused by their source and 12 failed. After the redistribution rule, 19,558 files from 25 websites, serving 74 entries.
The 2026-10-27 build ran ARB's raw_corpus build --redistributable-only again, in a container
with no network, over the first build's URL list and the added documents' fetch records (or,
for pages made from data, records written from their files); 24 URLs had no record
(the first build's failures), and every first-build document is unchanged except the re-dated
government tables.
- Downloads last month
- 86