kodebeat / papers / ai
Auditing code with a model, and knowing what was never read
A multi-pass LLM code audit with coverage accounting.
Runs several independent focused passes over each file, one class of defect at a time, then merges them with an agreement score. Every discovered file carries a recorded status and failed API calls are reported as failed tasks naming what they left unreviewed, so a clean report and an empty one are distinguishable. State lives in SQLite keyed by content hash, so a run resumes and an edit rescans one file.
What is in it
- The problem — what goes unanswered without it, and who notices first.
- Why the obvious alternative falls short — stated plainly, including where it is the better choice.
- How it works — the method, not a feature list.
- Concrete use cases — with console output quoted from the repository, never reconstructed.
- The methodology behind any number it emits — every term shown, so the figure survives a question.
- What it deliberately does not do — the section most papers leave out.
Part of the AI infrastructure theme.
Get the PDF
One email with the download link, and this paper already selected. No follow-up sequence.