The current UI was exercised in four disposable native suites: plan/file workflow (25 checks), editor/terminal/browser/recovery (51), general app/navigation/bots (97), and review component failure/cancellation (8). All 181 assertions passed. One compositor-only pointer-resize check was skipped because the virtual display has no window manager. The evidence gallery contains 55 native screenshots; tests establish behavior separately from the photographs.
tools/visual-evidence.mjs is an optional recorder used by the four existing smoke scripts. With SUPER_VISUAL_EVIDENCE_DIR unset, it is inert. With it set, the isolated native wrapper is required. Each suite writes its own PNGs and evidence.json: timestamped assertion names, capture hashes, executable hash, test-file hash, skip reasons, and completion state. An interrupted suite is marked incomplete and is not counted as passed. Use a fresh destination per run to preserve earlier receipts.
Set DEVELOPMENT_TEST_ROOT to a newly created disk-backed scratch directory with sufficient space. The system /tmp volume was nearly full during this session. Set SUPER_XVFB_BIN_DIR if Xvfb/xvfb-run are supplied from a local extracted tool directory. xauth must also be available. The wrapper isolates the display and process state; it does not restart the user's saved-world app.
Run these through bash tools/native-ui-test.sh node and retain stdout/stderr beside the receipt directories:
tools/development-plan-file-smoke.mjs
tools/development-smoke.mjs
tools/cockpit-app-smoke.mjs
tools/file-review-component-smoke.mjs
Use distinct APP_SMOKE_PORT values for simultaneous app suites. The workbench suite uses DEVELOPMENT_SMOKE_PORT and DEVELOPMENT_APP_PORT. Shell and provider fixtures bind loopback. Screenshot capture is scoped to windows descended from the test driver. Evidence-only scrolling places the asserted attachment/history content in view without changing product state.
The current task's outputs contain Super_UI_Evidence_Gallery.html, Super_UI_Change_Record.md, the full suite logs, original PNGs and receipt manifests. The gallery embeds its 55 photos and works as a portable offline artifact; the adjacent evidence folder contains the original images and logs. The change index relates the earlier implementation reports to the latest screenshots. Original successful captures are retained separately from the final captures with improved camera positions.
The report's images and links were checked directly. A live gallery browser preview could not be verified: the standalone browser failed to launch, and the browser tool blocked the local-file URL. No URL-policy workaround was used. Search/zoom interactions are not claimed as browser-tested. Markdown and original PNGs remain independently accessible.
Provider replies are deterministic local fixtures, not a real-model dogfood run. Some unavailable-state checks deliberately supply fixture state; the component suite uses explicit verification callbacks. Other integration checks exercise real native repository matching, exact file bytes, saves and recovery. No source or validation authority is inferred from screenshots.
Visual follow-ups: long editor error text truncates; narrow navigation is tall; isolated-display captures can show transient scrollbar lines across dialogs. Cross-platform, full keyboard/screen-reader coverage, every responsive breakpoint and normal-compositor border dragging remain unverified.
Durable attempts, result-bound validation, explicit acceptance, and a real-model Super-on-Super cycle remain the product critical path. This pass adds the repeatable evidence workflow, not a new development-attempt implementation. No commit, push or deployment was performed.