pithkit · CV-IR

AI says "done." Half the work is actually missing.

CV-IR is a tri-state validator that checks whether a code tree actually satisfies a list of acceptance criteria — not by trusting the AI's self-report, but by running programmatic probes and a separate LLM judge whose citations must be literal substrings of the code. Verdicts come back as pass, fail, or honest indeterminate.

The headline case (go-chi/chi): I wrote an intent that said "this router doesn't bundle authentication." The validator returned FAIL with cited evidence — middleware/basic_auth.go exists. The intent's claim was wrong about the code I had been pointing the validator at. CV-IR catches mismatches between what the intent claims and what the code does — the kind of overclaim no self-reporting AI can flag.

Try it yourself

Type an intent. Watch CV-IR author criteria and verify a real codebase.

Describe what a piece of software is supposed to do or be. An LLM blind to the target codebase proposes 5 acceptance criteria. The validator then runs probes + a separately-prompted LLM judge against one of four demo repositories. Verdicts come back with substring-verified citations. Total round trip: ~60 seconds.

Quickstart:
3 runs / IP / day · $5/day total budget
Author criteria from intent
Run validator on selected repo
Render verdict

Pre-baked case studies

Loading case data…