Datasets:
The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: ValueError
Message: Expected object or value
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 291, in _generate_tables
pa_table = paj.read_json(
io.BytesIO(batch), read_options=paj.ReadOptions(block_size=block_size)
)
File "pyarrow/_json.pyx", line 364, 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 string 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 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 544, 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 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 306, 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: Expected object or valueNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Natural-error yardstick for critics (v1)
What this is. 95 answers written by Llama 3.1 8B to explanatory questions in twelve topic areas. Each answer is labelled for real mistakes (a false fact, a wrong number or unit, or a reasoning step that does not follow) by two blind Claude passes (Claude Opus 5.5, and four Claude Sonnet 5.5 agents), with the disagreements adjudicated.
| Answers with an error | 59 |
| Clean answers | 36 |
| Error sentences marked by both passes | 84 (63 clear) |
| Labeller agreement | item agreement 0.82, Cohen's kappa 0.63 |
| Correction pairs | 31 (the original answer, and a copy with its agreed clear errors rewritten) |
| Screen judges | gemma4:31b, mistral-small:24b, granite4.1:30b, muse-glimmer, nemotron-3.5-lightning |
What it is for. Measuring how well a critic, or an LLM judge, finds errors that a model made while answering, as opposed to errors planted afterwards by an editing model. The correction pairs reverse the usual set-up: the edited copy is the better one, so a critic that only notices edits fails them.
Read before use
- The labels are made by models, not by people. They are useful ground truth for comparing judges, but they are not human-verified.
- Every answer comes from one generator (Llama 3.1 8B). A judge from the same family, such as muse-glimmer from Meta, may behave differently on it than on another generator's answers.
- In the v1 screens a failed judge reply is stored as an empty list. muse-glimmer had 1 such reply and nemotron-3.5-lightning had 39, so its v1 run is unreliable. A context-rich re-screen that records the reason for each failure will replace these in a later version.
Files. data/questions.json, data/answers.json, data/labels_claude.json,
data/labels_sonnet.json, data/gold.json (adjudicated), data/corrections.json,
data/selection.json, data/screen.json and data/screen_extra.json (judge flags), plus the
build scripts and the experiment README. datapack.json lists every file's SHA-256 and the model
pins. The pack files sit under files/. Check a download with rnd datapack verify natural-errors --copy <download>/files
(pip install mcptoolshop-rnd).
Licences
This dataset mixes material under different terms, file by file (see datapack.json):
- Answers, and any file quoting them (
answers.json,corrections.json,screen*.json): outputs of Llama 3.1 8B. Built with Llama. Llama 3.1 is licensed under the Llama 3.1 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Use is subject to the Llama 3.1 Community License and the Acceptable Use Policy. - Questions: outputs of Mistral Small (an Apache-2.0 model).
- Judge flags: outputs of Apache-2.0 models (gemma4, mistral-small, granite4.1, muse-glimmer) and of nemotron-3.5-lightning (NVIDIA Open Model License), quoting answer sentences.
- Labels, adjudication, correction sentences, scripts and README: MIT, © mcp-tool-shop.
- Parts were written by AI models (Claude, Llama, Mistral, Gemma, Granite, Muse, Nemotron), and none of the data is presented as human-written.
Source: mcp-tool-shop-org/rnd, experiments/natural-errors/.
- Downloads last month
- 7