The dataset viewer is not available for this split.
Error code: FeaturesError
Exception: ArrowInvalid
Message: Schema at index 1 was different:
model: string
None: double
genuine_empathic: double
negotiated_acceptance: double
performative_remorse: double
vs
feature: string
F_stat: double
p_value: double
sig: string
eta_squared: double
KW_H: double
KW_p: double
max_diff_models: string
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 244, in compute_first_rows_from_streaming_response
iterable_dataset = iterable_dataset._resolve_features()
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 4645, in _resolve_features
features = _infer_features_from_batch(self.with_format(None)._head())
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2866, in _head
return next(iter(self.iter(batch_size=n)))
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3073, in iter
for key, pa_table in ex_iterable.iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2581, in _iter_arrow
yield from self.ex_iterable._iter_arrow()
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 606, in _iter_arrow
yield new_key, pa.Table.from_batches(chunks_buffer)
~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
File "pyarrow/table.pxi", line 5012, in pyarrow.lib.Table.from_batches
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: Schema at index 1 was different:
model: string
None: double
genuine_empathic: double
negotiated_acceptance: double
performative_remorse: double
vs
feature: string
F_stat: double
p_value: double
sig: string
eta_squared: double
KW_H: double
KW_p: double
max_diff_models: stringNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Abbas Hamidavi
Department of Computer Engineering, Mashhad Branch, Islamic Azad University, Mashhad, Iran · ORCID 0009-0001-7342-7844
Moral Inversion Under Role-Play: Data and Materials
This dataset is also available as a GitHub repository with the same curated materials: https://github.com/AVABBAS/moral-inversion-under-role-play
Data and materials for the article:
Hamidavi, A. Moral Inversion Under Role-Play in Large Language Models. Submitted to AI and Society, November 2026. Preprint: Research Square, rs-11498569. The same materials are also deposited on the Open Science Framework: osf.io/uaf7h.
📖 About the study
Three frontier models (GPT-5.4, Claude 4.6 Sonnet, and Gemini 3.1 Pro) were run through a two-phase role-play design: an empathic baseline role first, and then an inverted role-play persona, with their moral responses tracked across phases and across transmission chains in which role content is passed onward. The models diverged sharply: two inverted substantially under the role-play, while the third refused in the large majority of its turns. Responses were coded against per-model codebooks, with a second rater's annotations and an agreement summary deposited alongside.
🗂️ Contents
| Folder | Contents |
|---|---|
transcripts/ |
The complete experiment transcripts for the three models (JSON). Full-render PDF versions are in the OSF deposit. |
codebooks/ |
The per-model codebooks (JSON). PDF renderings are in the OSF deposit. |
ratings/ |
The inter-rater agreement summary and the second rater's annotations. |
data/ |
The retained dataset tables: parsed experiment data (final and processed), the full dataset with features, extracted features and feature metadata, transmission fidelity, chain configurations and chain drift, phase comparison, acceptance and strategy by model, authority profiles, resistance summary, and the statistical summaries (ANOVA, effect sizes, significant correlations). |
code/ |
The feature-shift computation script and its results file. |
documentation/ |
The readme deposited with the OSF materials. |
Curation note
The OSF deposit also contains an earlier automated analysis layer: embedding projections (UMAP and t-SNE), cluster assignments and profiles, moral-oscillation tables, the notebook that produced them, and the automated reports written at that stage (one under the study's earlier working title). The submitted article does not rest on that layer, and its outputs are not consistent with the article's analysis, so they are not reproduced here. They remain available in the OSF deposit for the record. Figure images are likewise omitted in this text deposit; every plotted value from the retained analysis is present in the tables.
📚 Citation
Please cite the article. Citation metadata is provided in CITATION.cff.
🔗 Access to the article
- Preprint (Research Square, open access): researchsquare.com/article/rs-11498569
- OSF deposit: osf.io/uaf7h
- Journal: submitted to AI and Society; the publisher link will be added here if the article is published.
⚖️ Licences
Data, transcripts, and documentation are released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0, see LICENSE). Code is released under the MIT licence (see LICENSE-CODE).
The transcripts are outputs of publicly available AI systems, generated for this study in response to fictional scenarios. They contain no personal data.
- Downloads last month
- 128