Taming LLMs: Using Executable Oracles to Prevent Bad Code — Zero-Degrees-of-Freedom Programming Methodology

John Regehr / Hacker News·low signal

John Regehr published a methodology called 'zero degrees of freedom programming' where every LLM-generated code change is immediately validated against executable oracles (test suites, type checkers, sanitizers, formal specs) before acceptance. The approach constrains the LLM to only produce changes that pass all automated checks, eliminating the 'looks right but is subtly wrong' failure mode. 33 points, 18 comments on HN — practical framework for anyone shipping AI-generated code.