What an "unchecked" validation result means

When your coding agent validates code with preset AI, the result comes back with one of four verdicts. Three of them describe what the rules found. The fourth, unchecked, describes something different: there were no rules that could have found anything.

Copy link to sectionThe four verdicts

Verdict	What it means	Safe to write?
Blocked	At least one rule at the error level fired.	No. Fix it first.
Warned	Only warnings fired.	Yes, but review them.
Passed	Your design system's binding rules applied to this code and none of them fired.	Yes.
Unchecked	Nothing fired, but none of your design system's binding rules applied to this code.	Yes, but it is not a compliance pass.

Copy link to sectionWhy "unchecked" is not "passed"

A clean result is only meaningful if something was checking. A design system with no rules, or with rules that are all advisory, will never flag anything, so every piece of code would come back clean. Reporting that as "passed" would tell your agent, and you, that the code follows your design system when nothing actually looked.

So preset AI reports it as unchecked, and tells you how many rules applied: for example, "0 binding, 6 advisory".

Copy link to sectionWhat counts as a binding rule

A binding rule is one your design system has set to warn or block, and that applies to the code being checked:

• Forbidden primitives at warn or block that can apply to this code. A rule scoped to an element (for example, only inside a Dialog) applies only when that element is in the code.
• Context rules at warn or block whose file pattern covers the file and that set at least one constraint.
• Intent routes, when the agent says what the code is for and a route matches that intent.

Rules at suggest are advisory. They still show up as suggestions, but they never change the verdict, so they do not make a result "checked". Rules at observe are tracked silently and do not count either.

Checks that preset AI runs on its own (for example, hand-rolled component detection) still run, and anything they find still produces a warned or blocked verdict. They just do not turn a clean result into a pass.

Copy link to sectionWhere you will see it

• Your coding agent sees verdict: "unchecked" from validate_code, with a note that it is not a compliance pass. The write is allowed.
• The GitHub Action comment says no binding rules applied, instead of "All checks passed", and the unchecked-files output counts the files affected.
• The audit trail shows unchecked validations with a grey dot, never the green one.
• Compliance scores are left blank for unchecked code instead of reading 100.

Copy link to sectionHow to turn unchecked into checked

Give your design system rules at warn or block. The quickest start is the derived baseline: preset AI derives rules from your extracted tokens and applies the reliable ones at warn, so a design system with a real type, colour, spacing or radius scale is checked from the start. A design system whose tokens gave nothing to derive from stays unchecked until you add rules. See Derived baseline rules.

Copy link to sectionFAQ

Will an unchecked result fail my build? No. Unchecked never blocks. Only a human promoting a rule to block can make preset AI stop a write.

My design system has lots of rules. Why did a file come back unchecked? None of them applied to that file. Context rules only apply to files their pattern covers, scoped rules only apply when their element is present, and intent routes only apply when the agent's stated intent matches a route. Check the coverage field in the result to see which kind of rule is missing.