negprobe

Overview ¶

Command negprobe measures the false-confidence rate: queries about plausible concepts that do NOT exist in the indexed repo. A healthy stack flags these low-confidence (retrieval_health present); answering them with a confident-looking symbol list teaches the agent to trust junk.

This is the metric none of the positive-label suites can see: gen-eval, selfsweep and giteval all ask about things that exist.

Negatives are verified absent PER REPO before they count (see internal/negcases) — an unverified list measures the list, not the tool.

Usage:

negprobe -bin dist/contextmaxxer.exe -index path/to/index.db [-v]