The committed PNG baselines cover six desktop surfaces and two mobile screenshots. The protected production suite compares them with the repository's existing tolerance. Its 36 tests also preserve the navigation, flow, palette and accessibility checks. Three separate Turbopack development tests exercise the Windows service-detail regression with experimental.workerThreads: true.
Real mode uses different services, actual workflow IDs/timestamps, real policy output and measured local telemetry, so its screenshots are not interchangeable with seeded baseline data. Both were verified: deterministic screenshot comparisons preserve the accepted layout, while real-mode browser screenshots, overflow/keyboard assertions and accessibility scans exercise actual integration content.
| Surface | Real evidence | Difference classification / result |
|---|---|---|
| Developer Home | real-visual--home.png, browser--home.png |
EXPECTED DATA DIFFERENCE: actual owned service counts, latest operational attention and local greeting/time. Accepted structure preserved. |
| Catalog | real-visual--catalog.png, browser--catalog.png |
EXPECTED DATA DIFFERENCE: five generated services, real owners/health/Scorecards; no request history displays No samples. Same table/filter model. |
| Create Service | real-visual--create.png, browser--create.png |
INTENTIONAL INTEGRATION CHANGE: local platform description, supported Golden Path/runtime/plans and real workflow. Same nine-step UX and layout. |
| Service Detail | real-visual--services-checkout-api.png, browser--services-checkout-api.png |
EXPECTED DATA DIFFERENCE: real revisions, deployments, resources and health; tabs and dependency/Scorecard/runbook model retained. |
| Failed Deployment | real-visual--deployments-dep-75cb069f483c208c057f.png |
EXPECTED DATA DIFFERENCE: actual admission rejection/path and workflow ID. ACCIDENTAL REGRESSION FIXED: long technical paths previously widened grid content. Min-width/wrapping now bounds them. |
| Observability | real-visual--services-checkout-api-observability.png, latest production browser image |
INTENTIONAL INTEGRATION CHANGE: explicit local five-minute SLO window and unavailable/no-sample states; EXPECTED DATA DIFFERENCE: actual Prometheus series and Jaeger traces/logs. Application-only trace selection remains bounded. ACCIDENTAL REGRESSION FIXED: single CPU samples previously repeated the same time-axis label; now one point/timestamp appears. Empty series show No recorded samples and low request rates retain precision. |
| Mobile Home / failure | real-mobile--home.png, real-mobile--deployments-dep-75cb069f483c208c057f.png |
EXPECTED DATA DIFFERENCE: real service/version/policy content. ACCIDENTAL REGRESSION FIXED: real deployment header wrapped to avoid button overlap; no horizontal overflow. Historical fixture header retained for its accepted baseline. |
Images are retained under .gantry/evidence; the evidence ledger hashes the final comparison files. The above representative real surfaces were visually inspected. The real browser suite also scanned seven desktop routes and two mobile-critical routes with axe, finding no serious/critical violations. Code and pipeline overflow regions gained accessible keyboard focus. Command-palette focus enters the dialog and returns to its opener after Escape; tests cover this behavior. Status badges/text and actionable validation/remediation remain visible. This is bounded automated and keyboard/layout verification, not WCAG certification.
Additional intentional changes are local authentication/sign-out, server-permitted action gating, real SSE connection states, a collapsed deployment configuration control for admission demonstrations, and accurate read-only local-directory copy in team settings. They retain the Gantry typography, sidebar, dark-first surfaces, orange identity and information architecture.
The full real creation flow passed separately from the four other real-platform scenarios. Protected fixture E2E is not used as evidence that Kubernetes, policies or provisioning ran. See the final report and platform-browser-consumer.json, platform-e2e-recovery-final.json, platform-accessibility.json and browser-smoke.json for the actual integration evidence.