How this was made

This project was extracted and built by one person, in extended dialogue with several AI systems — Codex (OpenAI, GPT-5.5), Claude (Anthropic, Opus 4.8), and Fable. Given what the project is, I want to be precise about that, because “made with AI” now covers everything from honest collaboration to hiding the absence of work.

It was not used to write this for me. It was used to argue against me — and, more to the point, to help build an instrument that could catch me cheating even when the arguing failed.

The working method

The material of fallacy-cutter did not start as a product. It started as the methodology that produced a string of honest negative results on a much harder research question — the discipline of preregistration, controls, halt-on-fail, and tag-every-claim that I kept re-inventing in every experiment until it was worth turning into an instrument.

• Claude was the reasoning and writing partner — used to stress-test claims, find where an argument assumed its own conclusion, and push the prose until it said what I actually meant. Where this project is careful — “the instrument is real; the portable playbook is not finished” — that caution is usually the residue of an argument I lost to it first.
• Codex was the engineering partner — used to build and run the harness, extract this repository from the larger research forge without altering the source, and keep the worked example honest: a real leakage scan, a real preregistration, a real provenance signature.

The knife cuts its makers too

The point of the project is that trust should live in the instrument, not in the good intentions of whoever ran it — and that applies to this repository as much as to any experiment run through it. The worked example is reproducible from scratch and its decision is independently verifiable; the methodology docs are kept honest about what does not yet transfer. Neither model was treated as an authority, and neither was I: a claim survived only when it was still standing after repeated attempts to break it, and — for anything the harness governs — only when the harness signed off.

What stayed mine

The question was mine: can honest experimentation be made a property of the instrument rather than the experimenter? The constraints — fail closed, mark unprovenanced results non-citable, be honest about the unfinished part — were mine. And the responsibility for every claim here is mine: if something is wrong, it is wrong because I let it through, not because a model said it.

— Kirill Kruglov