Checking the work, and fixing what is wrong

Controls when each rule runs, set on the rule, on the workflow, or left to the product default — separately from how fixes are applied.

Picks out only the rules that matter for what is being checked, within a set budget, so the AI is never sent everything.

Checks that reach an answer without asking an AI anything, so they give the same result every time — the built-in check types, plus a sealed space for your own.

The AI-powered check, used only for questions a fixed check cannot answer, and always with the evidence behind its verdict.

An overall quality review sitting on top of the individual rule checks, with how deeply it looks set per workspace.

How something caught gets fixed — automatically, proposed for review, or only flagged — and checked again once a fix is applied.

Catches an AI rebuilding or contradicting something that already exists — by fixed checks first, and by judgment only where it has to be.