Catch drift from AI-generated UI
See where generated UI drifted, then decide whether to fix the code or improve the rule.
With a system connected, rules in place, and a tool installed, preset produces feedback as agents and people work. This is the payoff of the loop: you see when generated UI defers, and when it does not.
Validation, per snippet
When an agent generates UI, preset checks it against your rules right then. Each finding carries one of the three verdicts: forbidden (fix before it lands), suggested (a better path to consider), or resolved (already corrected). A clean check means the snippet conforms to the rules that exist.
Drift, over time
Drift is the slow divergence between a system as defined and a system as used. A token gets hardcoded in one place, then ten. No single change trips a rule, but the system erodes. preset surfaces this as a posture: where usage has pulled away, and by how much. Read it as a trend, not an alarm.
What to do with it
Every verdict resolves to one of two moves: fix the code, or change the rule. Making that call is the loop closing.
- Forbidden. Fix the code before merge. If the rule is wrong more often than right, loosen or retire it instead.
- Suggested. Weigh it. Apply when it improves the work; note why when you decline.
- Drift. Decide per case: reconcile the code back toward the system, or update the system where the new usage is the one that is actually right.
- Resolved or clean. Nothing to do.
The goal is not a green wall. It is a system whose definition and use stay close enough that what your agents generate is still recognisably yours. The full set of signals and how to tune them lives in the docs.