Open-Source Tool CauterRule v0.3.1 Fixes Zero-Pass Corpora With Data Repairs, Not Model Changes

CauterRule, an open-source sidecar that converts repeated AI agent failures into reusable rules, released version 0.3.1 after investigating two corpora that scored zero passes in the previous cycle. The 'adapters' corpus failed entirely because the reference pool contained no framework-specific trajectories, meaning the model was producing valid candidates that had nothing to be compared against. Similarly, the 'raw/ci' corpus suffered from four compounding data bugs, including a script that captured log headers instead of actual failure text and a reference pool padded with irrelevant trajectories. After fixing the underlying data issues — adding domain-matched reference signatures and correcting the log collector — the adapters corpus improved from 0/60 to 60/60, and raw/ci rose from 0/110 to roughly 21–26 out of 47 valid entries. The key finding is that a zero-percent corpus signals a data quality problem to investigate, not necessarily a model limitation to address.