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/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
return check_status(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 "/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/json/json.py", line 101, in _split_generators
pa_table = next(iter(self._generate_tables(**splits[0].gen_kwargs, allow_full_read=False)))[1]
~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
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 value
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.
Qualification by Calibration
Artifact and camera-ready materials for:
Carlos Toxtli and Manuel Delaflor. 2026. Qualification by Calibration: A Readable Benchmark for Admitting Language Models to Human-Computation Tasks. In 2026 ACM Conference on Human-AI Complementarity and Alignment (HCOMP 2026), September 27-30, 2026, Alexandria, VA, USA. https://doi.org/10.1145/3834580.3838735
The Trust-Calibration Benchmark (TCB) is a small, auditable qualification test for language models used as workers in human-computation pipelines: 60 yes/no items across five domains, each answered with a stated confidence, scored for accuracy, calibration (ECE, Brier), yes-rate, strict-format adherence, and availability under an audited administration protocol (template-guarded parsing, answers only from tagged or bare final lines, escalate-on-unparseable budgets, imputation flags). The paper's central finding concerns the screen itself: a naive harness (fixed token budget, unaudited last-match parser) manufactures failing workers out of competent ones, and the audited protocol is therefore part of the instrument.
Repository layout
code/ audited parser, runners, and offline analyzers (see code/README.md)
data/ the 60 items + gold labels, per-item label audit, per-trial
responses of all 9 models under all 3 protocols, repeated-sampling,
manipulation, and transfer-study data, cached LLM responses
tables/ LaTeX table fragments generated by the analyzers
figures/ figures generated by the analyzers (PDF + PNG)
paper/ camera-ready PDF, track-changes PDF vs the accepted submission,
CHANGES.md (how each review point was addressed), LaTeX source
Reproduce every number in the paper (offline, no model queries)
cd code
python3 analyze_panel.py # Table 1, reliability + forest figures, pairwise tests (Q1)
python3 analyze_harness.py # three-protocol harness comparison (Q2)
python3 analyze_classwise.py # confusion matrices, per-class P/R/F1, confidence resolution
python3 analyze_decouple.py # confidence-elicitation manipulation (Q3)
python3 analyze_criterion.py # transfer study: Spearman + permutation p + gate split (Q4)
python3 sensitivity.py # ECE with vs without confidence-imputed trials
python3 derive_naive.py # re-derive the naive condition from shipped raws (prints 0/540)
All analyzers are stdlib-only (matplotlib for figures) and read the shipped
per-trial data; only the run_*.py generators need a model endpoint.
Evaluate a new model or audit a new harness
python3 code/run_tcb.py <model-name> > my_trials.json
Set VLLM_BASE_URL and VLLM_API_KEY for your OpenAI-compatible endpoint.
License
Released under the MIT License (see LICENSE). The paper PDF in paper/ is
published by ACM under a CC BY 4.0 license.
Dataset card
| Paper | Qualification by Calibration: A Readable Benchmark for Admitting Language Models to Human-Computation Tasks |
| Venue | 2026 ACM Conference on Human-AI Complementarity and Alignment (HCOMP 2026) |
| Authors | Carlos Toxtli, Manuel Delaflor |
| License | MIT License (mit) |
| Dataset DOI | 10.5281/zenodo.24179040 |
| Version | 1.0.0 |
| Code repository | https://github.com/ai4he/qualification-by-calibration |
| Maintainer | Human-AI Empowerment Lab, Clemson University |
Dataset structure
| Path | Format | Size | Records |
|---|---|---|---|
code/ (15 files) |
md, py | 80.6 KB | |
data/ (17 files) |
json | 2.0 MB | 5,545 |
data/llm_cache/ (60 files) |
json | 14.1 KB | |
data/llm_cache_full/ (5700 files) |
json | 4.5 MB | |
figures/ (12 files) |
pdf, png | 479.3 KB | |
paper/ (3 files) |
md, pdf | 1.1 MB | |
paper/source/ (14 files) |
bbx, bib, bst, cbx, cls, dbx, pdf, tex | 373.5 KB | |
tables/ (2 files) |
tex | 2.3 KB |
5,823 files, 8.5 MB in total (metadata files excluded). Records are rows for CSV/TSV/JSONL files and list length for JSON arrays.
Known limitations
- Small by design (60 items); intended as a readable qualification screen, not a leaderboard.
Personal and sensitive information
None.
License
The data and documentation are released under the MIT License.
Citation
If you use this dataset, please cite the paper:
@inproceedings{Toxtli2026Qualification,
title = {Qualification by Calibration: A Readable Benchmark for Admitting Language Models to Human-Computation Tasks},
author = {Carlos Toxtli and Manuel Delaflor},
booktitle = {2026 ACM Conference on Human-AI Complementarity and Alignment (HCOMP 2026)},
year = {2026},
doi = {10.1145/3834580.3838735}
}
and the dataset:
@dataset{Toxtli2026TrustCalibrationBenchmarData,
author = {Toxtli, Carlos and Delaflor, Manuel},
title = {{Trust-Calibration Benchmark (TCB)}: Dataset for ``Qualification by Calibration: A Readable Benchmark for Admitting Language Models to Human-Computation Tasks''},
year = {2026},
version = {1.0.0},
publisher = {Zenodo},
doi = {10.5281/zenodo.24179040},
url = {https://doi.org/10.5281/zenodo.24179040}
}
- Downloads last month
- 36