FAIL CLOSED: Human-AI Research Without Pretending to Know

Summary

FAIL CLOSED by Alexanja Senke presents a method for using artificial intelligence in serious research without treating fluent output as truth. The book combines human structural intuition with machine-assisted reconstruction, computation, adversarial testing, and explicit governance over what is known, failed, or unresolved. Its central rule is that AI may be wrong, but it must not continue operating from an unresolved wrong state. The proposed architecture includes target-and-next-step checks, reconstruction before agreement, hidden-assumption audits, parallel-path verification, explicit claim boundaries, fail-closed research states, evidence maturity, weakest-link reasoning, and separation of public, protected, and unresolved work. The book describes case studies spanning mathematics, physics, and AI governance, including work related to totally positive matrices, Gaussian shift systems, exploratory Navier-Stokes and Riemann research, and execution-bound governance architectures. The outcomes are deliberately mixed: some investigations may produce a surviving theorem candidate, others a narrower claim, and others a clear rejection. The author presents failure as useful evidence rather than something to hide. The 172-page English Kindle book is aimed at independent researchers, engineers, developers, technical thinkers, founders, students, and others using AI for demanding intellectual work. It explicitly rejects the idea that AI replaces expertise, institutions, peer review, or human responsibility, and argues that disciplined boundaries and preserved uncertainty are essential to verifiable human-AI research.