The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ValueError
Message: Trailing data
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 290, in _generate_tables
pa_table = paj.read_json(
io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
)
File "pyarrow/_json.pyx", line 342, in pyarrow._json.read_json
File "pyarrow/error.pxi", line 155, in pyarrow.lib.pyarrow_internal_check_status
File "pyarrow/error.pxi", line 92, in pyarrow.lib.check_status
raise convert_status(status)
pyarrow.lib.ArrowInvalid: JSON parse error: Column() changed from object to number in row 0
During handling of the above exception, another exception occurred:
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 2840, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 506, in __iter__
yield from self.ex_iterable
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 398, in __iter__
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 304, in _generate_tables
batch = json_encode_fields_in_json_lines(original_batch, json_field_paths)
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 111, in json_encode_fields_in_json_lines
examples = [ujson_loads(line) for line in original_batch.splitlines()]
~~~~~~~~~~~^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/utils/json.py", line 20, in ujson_loads
return pd.io.json.ujson_loads(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
ValueError: Trailing dataNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
RCV-Bench: Research-Claim Verification Benchmark
Can an agent tell whether a claimed research result is actually supported? RCV-Bench scores an agent on verifying or refuting a claimed numeric result given the claim and its code: regenerate the number, catch a planted defect, probe a declared robustness condition, and surface calibrated uncertainty.
The benchmark is built around one distinction: re-execution is not the same thing as verification.
Code + live leaderboard: https://github.com/GrobeStreet/rcv-bench
Verdict classes
REPRODUCED · DEVIATION · FABRICATED · FRAGILE. A verdict object includes the verdict, the
regenerated value when available, a localized cause, a confidence, an escalation flag, and evidence.
Corpus — 14 instances across 4 source repositories
Spread: REPRODUCED 5 · DEVIATION 4 · FRAGILE 3 · FABRICATED 2.
| source repo | instances | example |
|---|---|---|
| de-stress-lab | 6 | 2.75σ time-variation falls to 1.81σ when LRG2 is removed → FRAGILE |
| arc-agi-2-occam-baseline | 3 | optimistic prefix analysis overturned by stricter same-holdout analysis → DEVIATION |
| FAIR Universe EXP-001 | 3 | external-gated weak-lensing cases used to test deviation/fabrication logic |
| mmlu-robustness-audit | 2 | ~78% underlying-answer flip rate under cyclic option reorder → FRAGILE |
Each task folder holds claim.json, a README.md, and a held-out GOLD/gold.json. Every gold label is
independently regenerable except the three FAIR tasks, which are external-gated (documented project records
used for task logic, not independently verified gold — the canonical FAIR repo is the authority).
Scoring
Verdict accuracy · macro-F1 across the four classes · defect localization · calibration (Brier/ECE) ·
escalation appropriateness. See scoring/.
Reference baselines
| system | verdict acc | macro-F1 | FRAGILE recall | FABRICATED recall |
|---|---|---|---|---|
| always-REPRODUCED | 5/14 (0.357) | 0.132 | 0/3 | 0/2 |
| naive-rerun | 9/14 (0.643) | 0.392 | 0/3 | 0/2 |
| perfect ceiling | 14/14 (1.000) | 1.000 | 3/3 | 2/2 |
Naive re-execution looks competent overall but misses every FRAGILE and FABRICATED task.
First real-agent baseline
A zero-API-cost autonomous local agent (Ollama qwen2.5-coder:1.5b, ≤8 steps/task, gold withheld, network
blocked) scored 6/14 (0.429), macro-F1 0.339, FRAGILE recall 0/3, FABRICATED recall 1/2,
localization 0/9, Brier 0.266. Not a frontier claim — the first end-to-end demonstration that tool use
and autonomy do not by themselves produce robust claim-verification.
How to run
See the GitHub repo: python3 build_tasks_reconciled.py, then the baselines and scoring/score.py, and the
harness/ external-process runner for evaluating your own agent. Public gold exists for auditability, so
publishable agent results require an execution boundary that prevents reading the checkout/gold during a run.
Licensing
Task data and gold: CC BY 4.0 (LICENSE-DATA.md). Benchmark code: Apache-2.0 (LICENSE). Source repositories
retain their own licenses.
Citation
Morong, R. (2026). RCV-Bench: Research-Claim Verification Benchmark. https://github.com/GrobeStreet/rcv-bench
- Downloads last month
- 89