BOKBO — Best of K Bad Options: K-Sample Disagreement Tracks Perturbation, Not Uncertainty

Anya Singh, Cabrel Happi, Jai Relan, Varun Nair, Vidyut Baradwaj · 2026

Read BOKBO — Best of K Bad Options: K-Sample Disagreement Tracks Perturbation, Not Uncertainty by Anya Singh, Cabrel Happi, Jai Relan, Varun Nair, Vidyut Baradwaj (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-30660.txt.

Description

Singh and colleagues address a hazard in test-time scaling for VLA policies, which sample actions and execute the verifier's favorite. When every candidate is unsafe, the system executes a violation with no warning. Their BOKBO layer adds conformal abstention to K-sample inference, with distribution-free guarantees on violation rate, in global and per-task variants. Analysis exposes a failure: policy confidence and sample disagreement track the action-noise setting while barely matching violations. A violation predictor supports calibration, and per-task BOKBO raises the worst task's hold fraction.

Catalog record

CREATORAnya Singh, Cabrel Happi, Jai Relan, Varun Nair, Vidyut Baradwaj
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41828
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.30660

Browse the full library · About the Tower of Babel