LLMs can't justify their answers — this CLI forces them

What is wheat?

Wheat is a decision‑making framework that aims to turn fuzzy LLM output into traceable decisions. It has been reported that wheat runs as a command‑line tool inside Claude Code, where you research, prototype, stress‑test, and then compile your findings into a formal decision brief. The slogan is clear: compile your decisions like you compile code — contradictions get caught before they ship.

How it works

The workflow is refreshingly dev‑centric. You iterate on options and evidence, then a "compiler" steps in and validates the brief; allegedly it refuses to produce a recommendation built on internal contradictions. The idea is neat and simple: make the model justify each link in the reasoning chain, not just hand you a polished answer. Think linters and type checkers for judgments — catch the nonsense at build time.

Why it matters

Why should anyone care? Because people are tired of chasing hallucinations. This is about auditability and accountability — two buzzwords now backed by a practical tool. Treating decisions like code also slots into a broader trend: engineers demanding reproducibility and guardrails from AI, not just clever outputs. Relief, if it works, could be immediate for teams that need defensible choices.

Caveats

Don't get carried away. It has been reported that wheat currently depends on Claude Code, so it's tied to whoever controls that stack — not exactly model‑agnostic. And a compiler is only as good as its rules: garbage in, garbage compiled. Still, the framing — force LLMs to justify, or don’t ship — is a crisp answer to a messy problem. Will it stick? Time will tell.

Sources: grainulation.com, Hacker News