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, and EU and Wikipedia texts under their licenses. The builder 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-price histories, consensus estimates and
third-party transcripts are among them, 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.
| file | docs | unpacked → archive | tree_sha256 |
|---|---|---|---|
L.tar.gz |
19,558 | 3.27 GB → 0.81 GB | 9319687f3fbed40a6c10ad8ba685cd2a6d28deff9125c7b6a879a07fc2564e47 |
How it was built
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.
- Downloads last month
- 24