# Third-party notices

This notice identifies third-party data and model sources. It does not assign a licence to the original companion code, notebooks, prose or figures. Source-specific notices in the extension folders remain part of this distribution.

## Human confidence data

`extensions/human/data.csv` is the unchanged 56,000-trial dataset supplied by Medha Shekhar and Dobromir Rahnev through The Confidence Database. Source files: [data](https://osf.io/download/mds32/) and [source README](https://osf.io/download/5ztpf/). The Confidence Database publication identifies its data as [CC0](https://creativecommons.org/publicdomain/zero/1.0/); see Rahnev et al. (2020), p. 318. The derived membership table and analyses are produced by the companion scripts; the original data file is preserved.

Please cite the data study and database when using these materials: Shekhar, M., & Rahnev, D. (2021), “The nature of metacognitive inefficiency in perceptual decision making,” *Psychological Review*, 128(1), 45–70, [doi:10.1037/rev0000249](https://doi.org/10.1037/rev0000249); and Rahnev, D., et al. (2020), “The Confidence Database,” *Nature Human Behaviour*, 4, 317–325, [doi:10.1038/s41562-019-0813-1](https://doi.org/10.1038/s41562-019-0813-1). These scholarly citations are requested by the source publication; CC0 itself imposes no attribution condition. The data dedication does not license the publishers’ article PDFs or figures, which are not included here.

## SWE-agent metadata extract

`extensions/agent/agent_runs.csv` is adapted from Nebius, [SWE-agent-trajectories](https://huggingface.co/datasets/nebius/SWE-agent-trajectories), revision `68195a1450865274106246d0d0296a1d6807b88e`, first shard `data/train-00000-of-00012.parquet`. The source dataset is licensed under [Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/). The extract contains all 6,670 rows of that shard. Changes comprise field selection and renaming, Boolean target encoding as 0/1, message-role counts, and retention of a source-row ordinal. It excludes trajectory text, repository code, patches and evaluation logs. Nebius has not endorsed this companion.

The source dataset card also directs users of underlying repository content and model outputs to the respective repository licences and Llama 3.1 terms. Consult those notices before retrieving or redistributing source material beyond the scalar metadata supplied here.

## Optional model probe

`extensions/agent/model_outputs.json` records four short arithmetic generations made with HuggingFaceTB/SmolLM2-360M-Instruct, revision `a10cc1512eabd3dde888204e902eca88bddb4951`. The [pinned model card](https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/blob/a10cc1512eabd3dde888204e902eca88bddb4951/README.md) declares [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). This is model provenance, not an assertion that generated outputs inherit that licence. No model weights, tokenizer assets or dependency-library source are distributed in this companion. Runtime settings and provenance are recorded beside the outputs.

Dependencies named in requirements files retain their own licences and are installed separately.