A Floor of 0.80 and a Ceiling of 0.63: The Semantic Channel That Never Fired
Update — v0.3.1 released. CauterRule is now live on GitHub and PyPI . It turns repeated agent failures into permanent standing rules — extract, replay-test, promote. pip install cauterule gives you the full CLI, framework adapters, rule lifecycle, pack ecosystem, and official rule packs. The v0.3.1 field test report evaluated 2 cloud models across 40 corpora and 4,742 trajectory-runs and is the…
A subtle change in the software triggered significant improvements. The new version, CauterRule, now works better at detecting paraphrases. Previously, a channel existed, but it never activated because it required a 0.80 threshold. This was too high, and low-scoring matches were ignored. The updated version lowered the floor to 0.62, aligning with the actual performance of the system.
The cosine similarity measure between paraphrases now falls between 0.63 and 0.80, showing genuine matches. With these changes, recall improved from 0.170 to 0.377, and the pass rate increased to 82%, exceeding the required 70% threshold. The updates were made without altering the model or prompts, focusing solely on the semantic channel and its floor setting.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
