The Behavioral Credibility Trilemma: When Calibrated Autonomy Becomes Impossible
Code and experiment data for the empirical validation of the Behavioral Credibility Trilemma via Best-of-N selection. Accompanies the manuscript (arXiv:2605.25739), prepared for submission to the Journal of Machine Learning Research (JMLR). v1.3 completes the public artifact set: per-model held-out calibration CSVs underlying the binding sets, the pre-specified cross-model replication (Gemma-2-9B, Yi-9B; all five hypotheses replicate 5/5 under per-model binding sets), the comonotone-selector robustness re-analysis, and the uniform-random and no-oracle-proxy control suites (plan amendment E-C).
Resource information
- Format
- zip
- Original repository
- 21440170
- DOI
- 10.5281/zenodo.21440170
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- mit-license
- Last record update
- Nov 5, 2026
About this resource
Code and experiment data for the empirical validation of the Behavioral Credibility Trilemma via Best-of-N selection. Accompanies the manuscript (arXiv:2605.25739), prepared for submission to the Journal of Machine Learning Research (JMLR). v1.3 completes the public artifact set: per-model held-out calibration CSVs underlying the binding sets, the pre-specified cross-model replication (Gemma-2-9B, Yi-9B; all five hypotheses replicate 5/5 under per-model binding sets), the comonotone-selector robustness re-analysis, and the uniform-random and no-oracle-proxy control suites (plan amendment E-C).