designesy.org/drift12 checkscompiled CSSstatic analysiscontract 0.1.1 provisional
Drift radar
Catch a UI drifting off its own system: token names that resolve to nothing, and values that stop clustering on a scale. Built for interfaces an AI model wrote in one session and edited in the next.
Every name the CSS uses is declared somewhere.
Token fabrication, d02 d11 d12: names that look like tokens and resolve to nothing.
Colors, spacing, radii, shadows and timing stay on a few steps.
Drift within a session, d03 to d10: one role, several raw values, in the same build.
Scan a URL and its grade lands on this scale. Point at any cell to read what it checks.
- Amnesia between sessions monitor
- Values worked out again from scratch, landing somewhere else.
- Silent breaking changes monitor
- Token values changed with no version bump.
A scan catches the two failure modes named under the cells. These two need history, which Monitor keeps.
- pass
- needs work
- fail
- does not apply
Method
Fetch the page
The HTML at the URL, then every stylesheet it links, plus inline style blocks.
Read the system
Every custom property declared, every var() reference, and each value set for color, spacing, radius, shadow and timing.
Run 12 checks
Four ask whether the tokens resolve. Eight ask whether the values cluster on a few steps.
Grade
Pass counts 1, warn 0.5, fail 0. Skipped checks leave the count. A is 90 and up.
score = (pass + warn × 0.5) ÷ checks scored × 100A ≥ 90B ≥ 80C ≥ 70D ≥ 60F belowStatic analysis of compiled CSS: no browser, no login.