Every design system drifts, and the fix isn’t a cleanup — it’s a repeatable audit.
A token file doesn't stay accurate on its own. Someone hits a deadline, hardcodes one hex value instead of reaching for a token, and ships it. Nobody notices, because it renders identically to the token-based version next to it. Six months later that same raw value has been copy-pasted into a dozen components, and the system has quietly split into "the tokens" and "the stuff that looks like tokens but isn't."
The audit is simple: grep every raw value in the codebase — hex colors, pixel values, opacity numbers — and check each one against the token file. If it matches an existing token, it's kept as-is. If it doesn't match anything, it gets flagged. That's the entire method. No new tooling, no dashboard, just a search pattern and a decision rule applied consistently.
The part people get wrong is treating a flag as a failure. A flagged value isn't a mistake the audit uncovered — it's the audit doing exactly what it was built to do. The alternative is a raw value sitting silently in production, invisible until someone builds on top of it and inherits the drift without knowing it's there.
Run the audit on a schedule — every release, every sprint, whatever cadence matches how fast the codebase moves — and drift never gets the chance to compound into a rewrite. Catching one hardcoded value costs five minutes. Catching forty of them after they've spread through the component library costs a lot more.

Leave a Reply