zero-doflisted

# Zero-DOF Programming Eliminate the freedom to do the job badly. Every measurable dimension of code quality gets an **executable oracle** — an automated tool that programmatically validates output. Dimensions without oracles get **human oversight checkpoints**. No dimension is left to LLM judgment alone. > Based on John Regehr's [Zero-Degree-of-Freedom LLM Coding](https://john.regehr.org/writing/zero_dof_programming.html). ## Quick Reference | Aspect | Detail | |--------|--------| | Purpose | Constrain LLM coding output through automated validation | | Core concept | Executable oracles — tools that programmatically validate code | | Key insight | Opposing oracles create the tightest constraints | | Playbook rule | Linear mandatory steps — no discretion, no shortcuts | | Gaming defense | Multiple opposing goals make gaming difficult or impossible | | Human zones | Architecture, unnecessary complexity, GUI polish, security | ## When to Use - Substantial feature work where correctness matters - Setting up a project for LLM-assisted development - LLM-generated code has recurring quality issues - Establishing automated validation for agent workflows - Designing playbooks or runbooks for coding agents - Reviewing whether agent output is properly constrained ## When NOT to Use - Trivial single-file edits with no quality risk - Exploratory prototyping where speed matters more than correctness - Tasks where the human is writing all code themselves - Pure documentation or conf