Validation & results

Frontguard's value depends on the AI correctly classifying visual changes. Accuracy is measured with two validation harnesses and tracked over time — not asserted.

Accuracy stats

395

tests across 26 test files in the core CLI

synthetic ground-truth cases with known classifications

PR-level

real-world validation against live GitHub pull requests

Synthetic validation

Ten programmatic before/after pairs with known ground truth, run against your own key:

$ npx tsx scripts/validate-ai.ts

Real-world validation

Validates against actual GitHub PRs through the full clone → render → diff → analyze pipeline:

$ npx tsx scripts/validate-ai-real.ts --repo shadcn-ui/ui --pr 1234