Sample record. Fictional demonstration content.
Abstract
Some tool failures return plausible but wrong output and raise no error. We detect such silent failures by comparing the distribution of tool outputs against a rolling reference and flagging shifts. On three deployed tools the detector finds 71% of injected faults with a 2% false-alarm rate.
Keywords silent failures · monitoring · tool outputs
Full text
CTM-PR-2026-0003.pdf
How to cite
Medina, C. & Mensah, K. (2026). Detecting Silent Tool Failures from Output Distributions. In Proceedings of the 1st Workshop on Reliable Tool-Using Agents (RTUA 2026) (pp. 19–24). CONTINUUM Proceedings. doi:10.xxxx/ctm.pr.2026.0003