The final verification reran the full suite in CI 43627688560 on the audited documentation head, with identical application code: 146 passed and one intentional skip. A fresh public HTTPS review also passed 42 checks across all seven routes and six size/theme combinations, plus 11 interactions covering filters, shortcuts, sorting, global/mobile search, empty recovery, wallet pickers and fully loaded signed-out support/disabled AI. No runtime errors or failed requests occurred. Existing captures and historical revision-specific checks below remain separate evidence.
Verified November 14, 2026: all 146 executed browser checks pass against the light-mode layout correction, using Playwright 1.65.1 and Chromium. GitHub run 43619439586 passed the full suite on published source 5c8dc527b3d6d29f6773ff6ac402c5886083f512, along with the application, contract and clean-container checks. Playwright discovers 147 scenarios; the desktop global-search scenario is intentionally skipped on mobile, which has its own queue search. The production build completed without warnings.
Local verification passes the production build, TypeScript, 30 relevant existing browser checks and three new short-screen regression checks. Actual appearance toggles were also inspected across 72 route/theme/viewport combinations, locally and on public HTTPS: nine sizes from 320 × 640 to 1440 × 960 on the queue, leaderboard, maintainer and support screens. Sidebar resource links remain reachable by keyboard within the viewport, including at 1024 × 600; 1280 × 720 fits without sidebar scrolling. The mobile footer and queue count retain readable line lengths.
The suite uses clearly labeled demo data. Wallet checks use the real selector modules and NEAR Connect sandbox with injected Nightly, popup-boundary and sandbox-executor fixtures; they do not approve a real account or sign transactions. Confirmation, support-session, ticket, signature callback and AI responses are explicit fixtures. The connected-header layout check substitutes public account markup only; it proves responsive layout, not wallet ownership. Real HTTP authentication and settlement evidence are recorded separately in COMMUNITY_E2E.json and LIVE_E2E.json.
- All main screens: work queue, bounty detail, posting/review, contributor profile, rankings, maintainer dashboard and support desk.
- Type/status/reward filtering, translation quick view, open-work quick view, conversation search and empty-result recovery.
- Loading skeletons, API retry, expired/settled work, disconnected wallet and terminal-state action hints.
- Exact storage/deposit review, draft preservation, pending/confirmed/failed transaction presentation and retained intent after mismatched evidence.
- Private support sign-in, appeal drafts, accessible/inaccessible threads, replies, resolution, expired sessions, NEP-413 callback validation and fragment scrubbing.
- Duplicate-funding acknowledgement, payout-cap disclosure, optional AI unavailable/failure/reply states.
- Deferred wallet-code loading, recovery despite a failed optional chunk download, and preservation of a pending action after a mismatched return hash.
- Four-provider connection dialog, account management, cancellation/retry, no-key MyNearWallet connection, Nightly account changes and complete NEP-413 challenge fields.
- Meteor/Intear sandbox network and identity requests, explicit Intear support limitations, and refusal to sign while a wallet switch is unfinished.
- Startup/cancellation of the shipped Intear executable, stale signature rejection after extension account changes, and revocation of delayed session responses after disconnect.
- Reward-category totals excluding expired work, translation discovery resetting filters, shared global search, and persisted appearance selection.
- Keyboard skip navigation, visible focus, serious/critical Axe WCAG A/AA findings and page overflow.
- Short-height sidebar containment and keyboard scrolling in both appearances; stacked mobile footer copy and unsqueezed queue count at 320 px.
Main layouts were tested at 1440 × 960, 375 × 812, and dark desktop appearance. An additional connected-wallet layout regression checks 320 px and 375 px in each appearance: header elements cannot overlap, account/theme/disconnect targets are at least 44 px high, form values use at least 16 px text to avoid iOS focus zoom, and testnet identification remains visible.
The dApp revision uses a full-viewport shell with integrated navigation, charcoal workspace, black inset queue and a yellow translation panel. The user's follow-up removed the outer yellow canvas and all outer gutters/corner radii. Manrope carries headings and financial values; Source Sans carries body text. Escrow, category and locale summaries come from real queue data. The first visit uses dark appearance; a saved light choice persists. Short desktop screens compact and scroll the sidebar without spilling its bottom copy onto the page. Mobile navigation and category tiles scroll within their own regions, account controls wrap beneath the breadcrumb, footer copy stacks, and the comparative leaderboard retains a labeled, keyboard-focusable scroll region.
The capture library includes 90 demo/fixture PNGs. Main-route, provider-picker and connected-header captures were refreshed for the full-viewport follow-up; state-specific fixture captures record the earlier wallet revision. Each full-page capture scrolls to the top first. Native dialogs use viewport captures because their top-layer backdrop covers the viewport. The visual review covered all seven main routes, both desktop appearances, mobile layouts, 320 px layouts and wallet dialogs. The earlier hidden-label overflow and split mobile financial decimal fixes continue to pass.
Automated accessibility checks cannot replace a complete assistive-technology audit. Browser fixture results must not be presented as real escrow or provider activity. The separate screenshots/live/ directory records the public deployment and is updated independently during deployment verification.
| Screen | Desktop | Mobile | Dark |
|---|---|---|---|
| Work queue | View | View | View |
| Bounty detail | View | View | View |
| Post bounty | View | View | View |
| Contributor | View | View | View |
| Leaderboard | View | View | View |
| Maintainer dashboard | View | View | View |
| Support sign-in | View | View | View |
| Private conversation fixture | View | View | View |
| Appeal draft fixture | View | View | View |
| Duplicate and payout-cap fixture | View | View | View |
| Connected header at 375 px (layout fixture) | View | View | View |
| Wallet provider picker | View | View | View |
| Wallet account fixture | View | View | View |
| AI reply fixture | View | View | View |
Each appearance directory also contains review-bounty, empty-filter-results,
loading-queue, queue-error, expired-bounty, wallet-not-connected, and the
three clearly named transaction-*-fixture captures. Fixture text is visible in
the transaction screenshots so they cannot be mistaken for real payouts.
Run after npm run build:
npm run test:uiPlaywright starts a demo-mode server. An existing server must also report demo mode; the browser tests do not target live signing. For a separately started QA server, set PLAYWRIGHT_BASE_URL. Screenshots are written to docs/screenshots/ by appearance. Failure traces and the HTML report remain local artifacts.
The suite lives in tests/ui.spec.ts, tests/community.ui.spec.ts, tests/performance.ui.spec.ts, tests/wallet.ui.spec.ts, and tests/wallet-bridge.ui.spec.ts. Viewport, browser and server configuration is in playwright.config.ts. Repeated runs overwrite review captures. Run a single meaningful subset when investigating a regression, for example:
npx playwright test --grep 'narrow connected-wallet header'The same app source was deployed at https://gitsettle.89.167.109.196.sslip.io and all seven main routes loaded without browser runtime errors or page overflow. These captures show real testnet data and signed-out public pages, not mocked account sessions: queue, mobile queue, light queue at 1280 × 720, light queue at 320 × 640, paid translation, posting, leaderboard, contributor, maintainer, support, and provider picker. No provider was selected in the live picker check, and opening it downloaded no additional signing scripts. The refreshed captures were visually rechecked after deployment.