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 80, 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.
Harvey LAB data packet
Study material for the harvey-lab benchmark of Applied RSI Bench (ARB): the
documents a lawyer would read to do the work in
Harvey LAB's tasks (statutes,
regulations, agency guidance, court opinions, filings, model documents and
practice material), as plain text where possible. An ARB post-training agent gets
the packet read-only at /home/agent/workspace/data_packet/ before its model is
evaluated.
The packet holds no LAB scenarios, case files, task instructions or rubrics: the builder refuses every URL in LAB's own repository.
| file | docs | unpacked → archive | tree_sha256 |
|---|---|---|---|
L.tar.gz |
143,503 | 12.78 GB → 5.42 GB | 9735eb126317c5de2d6fc872e690ba3f329a5db9e81509441f2abc24d9f9993a |
How it was built
ARB's data_packet builder (map → scout → script → list → build), 2026-10-04:
- map: one agent (Claude Fable 5.1) read the benchmark and listed what its work needs: 49 domains, 759 entries; an entry is one source, such as "Delaware Code Title 8, Chapter 1".
- scout: one agent (Claude Sonnet 5.5) per entry found where the entry's documents live online.
- script: one agent (Claude Sonnet 5.5) per website wrote a script that lists the site's documents for an entry, most important first (304 sites).
- list: code kept up to 1,000 documents per entry: 155,680 URLs.
- build: code fetched each URL once (robots.txt honored, per-site rate limits), turned HTML and PDF into text, and hashed every file.
Of the 155,680 URLs, 143,545 gave a document. 42 of them differ from another only in case or a trailing slash and hold the same bytes, so the packet stores 143,503 files. 11,869 URLs were refused by their source (robots.txt, a 4xx answer, or a scanned PDF without text) and 266 failed. Every entry has at least one document. The 79 bulk archives the scouts found (for example CourtListener and the Caselaw Access Project) are not included.
| kind | docs | GB |
|---|---|---|
| text, including PDFs turned into text | 62,499 | 5.04 |
| HTML pages turned into text | 76,808 | 4.28 |
| XML | 2,059 | 0.23 |
other formats as fetched (.zip, .xlsx, .docx, .xls, .doc, …) |
2,179 | 3.24 |
Layout
PACKET.json docs, bytes, tree_sha256, and per file: path, sha256, bytes, url,
entries (the map entries it serves), meta
README.md the map's 49 domains and 759 entries, each with its document count
<host>/<slug>.txt one normalized document per file (other formats keep their extension)
tree_sha256 is the sha256 of the sorted lines "{path} {sha256}\n", one per
document (PACKET.json and README.md excluded).
Use
hf download junlinw/arb-harvey-lab-data-packet L.tar.gz --repo-type dataset --local-dir .
mkdir -p data_packet && tar -xzf L.tar.gz -C data_packet
ARB pins the exact commit it mounts in the benchmark's benchmark.toml; pass
--revision <commit> to get the same bytes.
Sources and licenses
The documents come from 1,095 websites, and their terms differ. The builder's license table covers the sites below; "unknown" means the table has no entry for the site, not that the text is free to reuse. Many documents are copyrighted, for example those from commercial legal publishers, standards bodies and arbitral institutions. The packet is shared for research on post-training agents. Check each publisher's terms before any other use. To have a document removed, open a discussion on this dataset.
| license (from the builder's table) | docs | GB |
|---|---|---|
| unknown | 84,700 | 7.06 |
| us-government-work | 34,174 | 2.05 |
| public-record (sec.gov) | 15,498 | 3.41 |
| government-edict | 8,842 | 0.18 |
| eu-reuse-with-attribution | 106 | 0.01 |
| open-government-licence-v3 | 105 | 0.02 |
| permission-required | 41 | 0.01 |
| naic-copyright | 26 | 0.01 |
| pdl-1.0 | 24 | 0.01 |
| reproduction-of-federal-law-order | 18 | 0.02 |
| official-text | 8 | < 0.01 |
| cc-by-4.0 | 3 | < 0.01 |
Counts are per fetched document (143,545). These are not official versions; check the publisher before relying on a text.
Earlier version
Commit e1d159fb
held two smaller tiers, S (941 docs) and L (21,367 docs), limited to the
official texts of the laws LAB's rubric lines cite.
- Downloads last month
- 76