test(ui): align single-choice picker locators

Author: waleedlatif1

apps/docs/COPY_AUDIT.md

@@ -691,7 +691,7 @@ Every row below represents a fully reviewed inventory entry. “Rewritten” inc
 - **API inputs and examples**: 22 canonical contract files reviewed: 19 editorial-stripped ASTs match staging; three differences are documented description-only helpers. All 135 rendered JSON request bodies validate, and all 135 curl samples pass bash syntax. The baseline’s 17 invalid bodies were reproduced and corrected.
 - **Generator boundary**: The actual 1,057,209-byte resources input reproduced the old successful empty-format result. The formatter now has explicit 16 MiB bounds, preserves complete parsed JSON (1,027,061 formatted bytes), and rejects oversized input; independently inspected.
 - **MDX and links**: All 790 runtime entries compiled/exported. All 1,988 internal links and 97 preserved inbound anchors pass, including excluded integrations and application references.
-- **Repository gates**: Root lint (26 tasks), type-check (26 tasks), check:audits (60 audits), test (20 workspace tasks plus script tests), and applicable block-registry checks against origin/staging passed. Full app run: 3,378 files, 37,636 passed, and 12 skipped. Existing warnings in two unchanged tests remain. No tests were added, edited, or removed.
+- **Repository gates**: Root lint (26 tasks), type-check (26 tasks), check:audits (60 audits), test (20 workspace tasks plus script tests), and applicable block-registry checks against origin/staging passed. Full app run: 3,378 files, 37,636 passed, and 12 skipped. Existing warnings in two unchanged tests remain. The initial docs audit added, edited, or removed no tests; the later picker-selector follow-up is recorded below.
 - **CLI discovery**: Existing tests reproduced a real search regression when the word spent was removed. Help now retains credits spent without the inaccurate this-period claim; all 17 search tests and the complete CLI suite pass.
 - **Production build**: Final apps/docs build passed after the source and layout corrections, generating 1,588 static outputs and full Markdown. A dummy loopback DATABASE_URL was used; no database or authenticated product request was exercised.
 - **HTTP checks**: 1,631/1,631 pass: 789 HTML routes, 789 per-page Markdown routes, 37 local assets, root/missing-route/noindex checks, full Markdown byte equality and API-operation exclusion, CLI.mdx, OpenAPI JSON, negotiated Markdown/Vary Accept, and integration landing backlinks.
@@ -701,11 +701,12 @@ Every row below represents a fully reviewed inventory entry. “Rewritten” inc
 - **Colors and preview**: All five HTTP method badges checked in light and dark themes; minimum computed contrast is 8.43:1. Inline prose code matches Sim Chat at normal weight, inherited font size and 1px/4px padding in both themes; fenced samples and classed API/schema code keep their own chrome. Dark appearance restored and temporary viewport reset; a verified production API preview remains available.
 - **Cleanup**: 111 newly unused original assets removed (15,187,771 bytes), with repository consumer proof. Unused ActionMedia and three workflow previews removed. Authored callouts 234→117, FAQ 58→0 with useful answers moved, media 178→23, and placeholders 36→0. Shared integration assets retained.
 - **Evidence**: Detailed local evidence in /tmp/sim-docs-copy-audit includes exact final-comparison/rewrite ledgers, source qualifications, rebase reviews, parity proofs, formatter reproduction, HTTP/render reports and screenshots. The committed audit retains every page’s final comparison and preservation findings.
-- **Ship cleanup**: Eight read-only cleanup passes found no React state/effect/memo/callback/query/URL/design issue. Three comment improvements place conditional-write documentation on its owner and shorten redundant reranking/retention notes. No tests changed. Public hygiene review cleared all 335 regex hits as synthetic examples, documented limits, public references, or repository measurements.
+- **Ship cleanup**: Eight read-only cleanup passes found no React state/effect/memo/callback/query/URL/design issue. Three comment improvements place conditional-write documentation on its owner and shorten redundant reranking/retention notes. That cleanup pass changed no tests. Public hygiene review cleared all 335 regex hits as synthetic examples, documented limits, public references, or repository measurements.
 - **Review corrections**: PR review triage corrected 13 findings and aligned three stale convention findings with sentence case; four findings were disproved against owners and real rendering/help output. A complete 303-operation sweep checked 539 canonical input slices and corrected 33 parameter/query metadata claims plus eight empty-query clarity descriptions. MCP probe/reset, hosting, billing source, credential capability, SSO save result, log/connector examples, numeric preview, and Pi color are corrected at their owners. Related request-cap guides distinguish configured caps, decoded-file limits, and effective request limits.
 - **Review-round validation**: All regenerated references, eight cleanup passes, root lint, 60 audits, 26 type-check tasks, block-registry checks, five sibling test files (36 tests), and the complete 20-workspace test run pass. App: 3,378 files, 37,636 passed, 12 skipped. Production docs build passed; 135 request examples and 135 curl samples validate. All 1,631 HTTP/Markdown/asset checks pass again. API headers fill mobile/tablet widths across 12 breakpoints, MCP clears its TOC at 320/768/1023px, and the temporary viewport is reset. Deployed Vercel preview is login-protected; content verification uses the local production build.
 - **Review-round UI validation**: Both later review findings are corrected: distinct Show/Hide array-item action labels and the canonical Pi canvas sentence with its repository field. Auth labels render without nested code pills, preserve keyboard selection, and keep 30px visible faces with 44px targets. Response headings align with the shared ChipSelect, and its 96px menu aligns to the trigger within 1px at 320/390/768/1023/1280px. Response and request-example pickers retain accessible labels and selection callbacks. Property filters use scoped dependency selectors, including body-portaled schema popovers; focus has no outward ring, narrow inputs fit their frame, and coarse-pointer mobile fields use the shared 44px/16px treatment.
 - **Review-round control coverage**: All 303 response menus preserve all 2,919 status choices in canonical order. All 22 choices across eight multi-example request pages produce distinct, nonempty code samples. Seventeen selected response states across list and execution pages show exactly their matching open schema. Matching and empty-result filtering, nested-schema Enter navigation, and Escape focus restoration pass. Fresh final 320px checks cover all 789 routes, including unchanged integrations, with zero root, prose or visible filter-input overflow; intentional code/table/tab/canvas scrolling is retained. All 1,631 HTTP/Markdown/asset checks pass after the final production build.
 - **Review-round repository gates**: The final UI changes pass eight cleanup passes, root lint (26 tasks), 60 audits, type-check (26 tasks), block-registry checks, all three docs test files (seven tests), and root tests (20 successful workspace tasks plus script tests; 19 unchanged workspace results reused). An earlier run failed on worker startup/termination and a teardown timeout; all eight affected files then passed (68 tests), and the complete rerun with two workers passed 3,378 app files, 37,636 tests, and 12 skips. No tests or timeout configuration were changed to suppress those failures. Final production docs build generates 1,588 outputs.
 - **Accessibility follow-up**: Request-example and response-status trigger names now include the selected value. Browser accessibility trees confirm changed selections announce Request example 3 and status 413 while callbacks, options and styles remain unchanged. Eight cleanup passes are clean. The later tablet TOC warning was independently disproved from compiled CSS: important utilities-layer declarations win over the unlayered important ID rule; fresh 320/768/1023px computations reserve 40px, with tablet headings 32px below the TOC. Full tests and production-build validation for this labels-only follow-up run in CI under the latest user instruction. The earlier all-page and control render proofs still describe the unchanged visual implementation. Normal docs dev startup adds the installed Next.js guidance to AGENTS.md; that standard section is retained alongside the authored editorial instructions.
 - **Open-menu accessibility and hydration**: The shared ChipSelect now exposes single choices as menuitemradio with aria-checked from its existing controlled value, matching the main app reasoning-effort picker. Radix keyboard handling, callbacks, row chrome and the multi-select checkbox branch are unchanged; consumer, role-selector and eight cleanup reviews found no behavioral regression. The pre-fix accessibility tree exposed plain menu items; final keyboard checks show exactly one checked choice, its state follows selection, and Escape restores trigger focus. All 22 request examples pass again at 320px with distinct code and correct checked state; response 200/413/503 checked states, schemas, 96px menu width and alignment pass. Docs pickers are nonmodal to avoid applying aria-hidden to the lazily hydrating search footer: the warning was reproduced, then three fresh-page menu checks passed with no warning or hidden siblings. One synthetic interaction before navigation focus settled was repeated after the page settled; no data action occurred. Full final tests and builds run in CI.
+- **Picker-test compatibility**: CI reproduced eight failures in four existing tests whose single-choice helpers queried the old menuitem role. Parallel test-audit review retained their shell-quoting, explicit integration-grant, scheduling and unsaved-draft assertions and changed only those helpers to menuitemradio. The existing mobile E2E settings-navigation locator received the same one-line adaptation; ordinary action and ChipDropdown locators remain unchanged. No tests, assertions or production seams were added or removed. Final full suites and mobile E2E run in CI; local regeneration, lint, 60 audits, 26 type-check tasks and block-registry checks pass.

apps/sim/app/o/[organizationId]/settings/components/integrations/search-source-setup.test.tsx

@@ -280,7 +280,7 @@ async function openSyncFrequency() {
 }
 
 function menuItem(label: string): HTMLElement {
-  const match = Array.from(document.querySelectorAll<HTMLElement>('[role="menuitem"]')).find(
+  const match = Array.from(document.querySelectorAll<HTMLElement>('[role="menuitemradio"]')).find(
     (node) => node.textContent?.trim() === label
   )
   expect(match, `Menu item ${label}`).toBeDefined()

apps/sim/app/o/[organizationId]/settings/components/search-mcp-connection.test.tsx

@@ -37,7 +37,7 @@ describe('Search MCP client connection', () => {
     await act(async () => {
       trigger.dispatchEvent(new KeyboardEvent('keydown', { key: 'Enter', bubbles: true }))
     })
-    const item = Array.from(document.querySelectorAll<HTMLElement>('[role="menuitem"]')).find(
+    const item = Array.from(document.querySelectorAll<HTMLElement>('[role="menuitemradio"]')).find(
       (element) => element.textContent?.trim() === label
     )
     expect(item, `Expected the ${label} client`).toBeDefined()

apps/sim/components/settings/unsaved-changes-regression.test.tsx

@@ -786,7 +786,7 @@ function select(label: string, optionText: string) {
   const trigger = container.querySelector<HTMLButtonElement>(`button[aria-label="${label}"]`)
   if (!trigger) throw new Error(`Missing select ${label}`)
   act(() => trigger.dispatchEvent(new KeyboardEvent('keydown', { key: 'Enter', bubbles: true })))
-  const option = [...document.querySelectorAll<HTMLElement>('[role="menuitem"]')].find(
+  const option = [...document.querySelectorAll<HTMLElement>('[role="menuitemradio"]')].find(
     (node) => node.textContent?.trim() === optionText
   )
   if (!option) throw new Error(`Missing option ${optionText}`)

apps/sim/ee/credential-groups/components/organization-workspace-grant-modal.test.tsx

@@ -83,7 +83,7 @@ describe('workspace integration grant editor', () => {
     await act(async () =>
       trigger.dispatchEvent(new MouseEvent('pointerdown', { bubbles: true, button: 0 }))
     )
-    const option = [...document.querySelectorAll<HTMLElement>('[role="menuitem"]')].find(
+    const option = [...document.querySelectorAll<HTMLElement>('[role="menuitemradio"]')].find(
       (element) => element.textContent === 'Finance'
     )
     expect(option).toBeDefined()

apps/sim/scripts/test-mobile-e2e.ts

@@ -1111,7 +1111,7 @@ async function exerciseViewport(
         await expectContained(page, navigation)
         await navigation.getByRole('button', { name: 'General', exact: true }).tap()
         await expectContained(page, page.getByRole('menu'))
-        await page.getByRole('menuitem', { name: 'General', exact: true }).tap()
+        await page.getByRole('menuitemradio', { name: 'General', exact: true }).tap()
         await expect(page.getByRole('menu')).toHaveCount(0)
         await expect(page).toHaveURL(/\/selfhost\/settings\/general/)
         await expect(navigation).toBeVisible()