Tokens as a single source of truth pipeline

WHAT THIS TESTS: This evaluates whether you can architect a token system with one authoritative source feeding many platforms automatically and reliably.

A GOOD ANSWER COVERS: Establish one source of truth: token definitions in a standardized, platform-neutral JSON format, ideally following the W3C Design Tokens Community Group spec so tooling interoperates. Structure them in tiers, primitive, semantic, and component, with semantic tokens referencing primitives. A build tool, commonly Style Dictionary, ingests the JSON, resolves aliases, applies platform-specific transforms for units, color formats, and naming conventions, and uses formatters to emit concrete outputs: CSS custom properties for web, a Swift file for iOS, and a Kotlin object or Android XML resource. Package and version each platform output with semantic versioning and publish to the appropriate registry so consumers pin and upgrade deliberately. Run the entire pipeline in CI triggered by changes to the source, with validation and visual regression checks, so a single token edit regenerates and republishes all platform artifacts consistently. Designers can author tokens via a plugin that exports the same JSON, keeping design and code aligned.

COMMON WRONG ANSWERS: Allowing multiple sources of truth, such as separate hand-maintained files per platform, which drift. Skipping the transform stage so units and colors are wrong on some platforms. Not versioning or automating, forcing manual republishing. Ignoring alias resolution between semantic and primitive tiers.

LIKELY FOLLOW-UPS: Why standardize on the W3C token format. How do designers feed tokens into the source without breaking it. How do you handle composite tokens like typography and shadows across platforms.

ONE CONCRETE EXAMPLE: A color.action.primary semantic token aliases color.blue.500. On commit, CI runs Style Dictionary, emits --color-action-primary for web, a Swift constant for iOS, and an XML color for Android, bumps the package versions, and publishes, so all three platforms get the same decision from one edit.