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.
CausalSmith run records
Raw records of the research runs behind CausalSmith, an agentic pipeline that proposes causal-inference theorems, derives them informally, and formalizes and proves them in Lean 4 on top of the Causalean library. The papers, their Lean proofs and the website are in the GitHub repository; this dataset holds what a run leaves behind on the way there, for accepted runs and for the ones that were shelved.
Layout
One compressed archive per unit, <group>/<id>.tar.zst:
| Group | A unit is | Unpacks to (inside a clone of the repository) |
|---|---|---|
accepted |
a run whose result was accepted | CausalSmith/doc/research/_bank/accepted/<id>/ |
downgraded |
a run banked below its target tier | CausalSmith/doc/research/_bank/downgraded/<id>/ |
failed |
a run that did not reach a result | CausalSmith/doc/research/_bank/failed/<id>/ |
study |
a study run that built reusable Lean substrate | CausalSmith/doc/study/<slug>/ |
topics |
a topic-selection round | CausalSmith/doc/research/_topics/... |
manifest.jsonl lists every archive with its sha256, size and file count.
A run's archive typically holds its state file and README, the proposal and derivation cores,
the derivation history, referee reports, the orchestrator's decision log, and the saved output of
each model call (logs/stages/). A topics archive holds slates, literature and collision checks,
gate verdicts and presolve receipts.
Getting the records
From a clone of the repository:
scripts/fetch_runs.sh --list
scripts/fetch_runs.sh failed/<id>
scripts/fetch_runs.sh accepted
or directly, e.g. hf download jytan12/causalsmith-runs --repo-type dataset --include "accepted/*",
then tar --use-compress-program="zstd -d" -xf <file>.
What was changed or left out
- Machine paths, logins and account addresses are replaced by placeholders such as
<repo-root>,<home>and<user>. - Lean sources are not included; the published proofs are in the GitHub repository.
- Papers, source bundles, bibliographies and software the agents fetched while working, and text extracted from them, are removed by rule (file type, location, and markers such as an author line or a copyright notice). The agents' own notes and search records remain; they cite and may quote the work they discuss. If you find a third-party document that the rules missed, please report it on the GitHub repository.
- Access tokens in recorded links (signed download links, session parameters) are redacted.
- TeX compile transcripts and other regenerable byproducts are not included.
These are unedited working records of language-model agents. They contain wrong turns, refuted claims and referee objections; a statement in a log is not a verified result. The verified results are the Lean theorems in the repository.
- Downloads last month
- 36