The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ArrowInvalid
Message: JSON parse error: Invalid value. in row 0
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 324, in _generate_tables
df = pandas_read_json(f)
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 38, in pandas_read_json
return pd.read_json(path_or_buf, **kwargs)
~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 815, in read_json
return json_reader.read()
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1014, in read
obj = self._get_object_parser(self.data)
File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1040, in _get_object_parser
obj = FrameParser(json, **kwargs).parse()
File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1176, in parse
self._parse()
~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/pandas/io/json/_json.py", line 1392, in _parse
ujson_loads(json, precise_float=self.precise_float), dtype=None
~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: Expected object or value
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 489, 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 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
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 327, in _generate_tables
raise e
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: Invalid value. in row 0Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ScienceMysteryBench — hosted sessions
Raw model sessions behind the manifest fields that
ScienceMysteryBench
declines to commit: oneshot_codex_transcript, oneshot_claude_transcript,
memory_probe_transcript. CONTRIBUTING is explicit that these are hosted and
never committed, and that a reviewer must be able to open them.
Gated on purpose. Every file here pairs the scrubbed data with the correct
answer, which is exactly the combination the canary line exists to keep out of
training corpora. Gating costs a reviewer one click and costs a crawler the
whole directory. This is a separate repository from
sciencemysterybench-data for a hard reason: that one is fetched anonymously
at Docker build time by every task, so it cannot be gated.
oneshot-codex/
One-shot runs under the harness PR #26 names as the Codex half of the
difficulty floor — Codex / GPT-5.6 Sol / xhigh. A task must defeat this
configuration to leave draft.
| slug | reward | wall clock | budget used | error vs tolerance |
|---|---|---|---|---|
| sm-0002 | 1.0 | 21m26s | 18% of 7200 s | 12 s of ±120 s |
| sm-0003 | 1.0 | 14m19s | 24% of 3600 s | exact match |
| sm-0006 | 1.0 | 7m45s | 6.5% of 7200 s | 0.09 of ±3.0 |
All three passed. None is a near miss: time budgets barely touched, numeric
errors far inside tolerance. oneshot_codex_failure cannot be filled in
truthfully for any of them.
Per directory: codex.jsonl is the raw codex exec --json stream,
run.json the configuration and cost, reward.json and
answer_record.json the verifier's own output, trial.log the harness log
including the exact agent invocation.
memory-probe-reviewer/
The reviewer-side measurement CONTRIBUTING requires be published beside the author's and never averaged.
- sm-0003 —
NAOw/decrease, reward 0.0. Independently reproduces the trap the author's Opus 5 probe fell into: both models reason that 6 of 12 columns are NAO variants, so the study must be about the winter NAO. Two models, two wrong answers, same wrong column. - sm-0002 — the
source_classpart alone, with no data: correct, in 39 reasoning tokens. The author already recorded this limitation; this measures how large it is. The lesson generalises: probe multi-part keys part by part, because the whole-answer probe scores 0.0 and hides a free part.
Reproducing
harbor run -p tasks/sm-NNNN -a codex -m gpt-5.6-sol \
--ak reasoning_effort=xhigh -o ~/smb-runs -q -y
run.json records the run-side allowlist additions each trial needed (the
task packages were not modified) and the egress facts verified from inside the
agent container: huggingface.co, doi.org, datadryad.org and
api.crossref.org were all blocked.
- Downloads last month
- 165