Accessibility result: Studio and Operator surfaces

Walked: 2026-05-29 across the then-current 483 Studio routes and 10 Operator routes; the repair record was extended with full reruns on 2026-05-30.
Reconciled: 2026-07-17 against the current route tree, shared axe fixture, dynamic Studio render smoke, representative Studio accessibility matrix, and focused Operator suites. This was a source reconciliation, not a rerun of the dated full sweeps.
Verdict: partial — the 2026 sweep remains valuable remediation evidence. Current automation renders every discoverable static Studio route, runs the serious/critical axe gate on 39 representative Studio routes, and exercises important Operator surfaces through the same gate. It does not provide a current all-route, all-impact, or manual assistive-technology sign-off for either surface.
Historical runner: node /tmp/walk/at-axe-parallel.mjs <studio|operator> <concurrency>, which wrote /tmp/walk/results-axe-<surface>.jsonl. These temporary paths are provenance only; the repository suites below are the portable evidence.

Result at a glance#

Evidence lane	Recorded or current contract	Authority limit
2026-05-29 Studio sweep	483/483 then-current routes rendered; the initial audit exposed clustered workspace-pattern debt	Historical build, inventory, runner, and raw report
2026-05-30 Studio remediation	Full 483-route rerun recorded zero axe violations and one h1 per route after shared-component repairs	Dated result, not today's maintained matrix
2026-05-29/30 Operator sweep	10/10 then-current routes recorded zero axe violations plus one main and h1 after repairs	Historical route set
Current Studio render contract	App-router discovery currently finds 479 static routes; each must authenticate and expose a heading	Most tests explicitly skip axe
Current Studio a11y contract	39 representative routes run the WCAG 2.1 A/AA serious/critical axe gate	Not all 479 routes; moderate/minor do not fail
Current Operator contract	Hub, tenant, tenants, incidents, and editorial-review behavior run through focused specs and the shared axe fixture	Representative flows across a 13-route static tree
Manual assistive-technology pass	Not performed by these automated records	No VoiceOver, TalkBack, NVDA, switch, voice, or AT QA

Evidence map#

The dated audit and today's maintainable automation answer different questions. The current Studio suite keeps full render/auth breadth, then spends axe cost on a representative route matrix; Operator depth is organized around key workflows.

flowchart LR A[2026 full Studio and Operator sweeps] --> B[Shared-component remediation] B --> C[Dated zero-violation records] D[Current Studio route tree] --> E[479-route render and auth smoke] D --> F[39-route representative axe matrix] G[Current Operator route tree] --> H[Focused hub, tenant, incident, and review suites] F --> I[Serious and critical WCAG gate] H --> I C -. historical evidence only .-> J[Current release claim] E -. render is not axe .-> J I --> J K[Manual assistive-technology QA] -. still required .-> J

Proven observations#

Historical sweep and repair campaign#

The first clean Studio rerun rendered real authenticated workspace content on all 483 then-current routes. The audit found roughly 1,900 violation nodes, concentrated in reusable card, control, form-field, heading, role-list, scroll-region, and entitlement-interstitial patterns rather than 483 unique defects.
The repair log records elimination of nested interactive controls, undersized targets, missing names, contrast failures, invalid ARIA relationships, non-focusable scroll regions, multiple h1s, and the incomplete Isis interstitial shell. The final 2026-05-30 note reports zero axe violations and zero routes with an h1 count other than one across that dated 483-route set.
Operator repairs raised low-contrast status text and restored a main landmark and h1 on affected surfaces. The dated reruns recorded 10/10 routes clean.

Current durable Studio contract#

studio-route-render-smoke.spec.ts discovers static routes directly from src/app/studio at collection time, rejects implausibly small inventories, and currently enumerates 479 routes. Dynamic segments are excluded because they need seeded parameters and dedicated specs.
Every discovered route must retain authentication, settle inside the shell, and expose a visible top-level heading. Routes outside the accessibility matrix explicitly use skipAxe: true.
The 39-route A11Y_MATRIX samples the Studio root and named groups such as Aja, Bellona, Hathor, Isis, Neith, and Yemaya. Those tests inherit the shared WCAG 2.1 A/AA scan and fail on serious or critical findings.

Current durable Operator contract#

The current app tree contains 13 static Operator pages. The focused operator-surfaces suite covers the hub fan-out, stable surface targets, mobile collapse, incident severity, tenant directory, tenant console, local dialogs, and anonymous gating.
The editorial-review suite separately proves the /operator/studio queue, selected artifact, transcript/waveform, notes, citations, decision entry point, static-row semantics, and anonymous gating.
Both specs import the shared fixture, so successful tests finish with the serious/critical WCAG scan unless they explicitly opt out. They are deep workflow evidence, not an automatically enumerated 13-route audit.

Boundaries and gaps#

Historical full sweep is not the current durable matrix. The 483/10 route zeros belong to the 2026-05-30 build and temporary runner. Route inventories, DOM states, axe versions, and product behavior can change after that date.
Render coverage is not axe coverage. The current Studio suite covers authentication and heading render across 479 static routes, but only 39 representative routes receive its automatic axe scan.
Operator coverage is representative. Focused suites deeply exercise key Operator workflows; no current collector proves every one of the 13 static pages at every viewport and state.
The maintained gate is serious/critical, not zero all severities. The shared fixture scans WCAG 2.1 A/AA tags and fails on serious or critical impact by default. Moderate and minor results require separate review or a stricter failOn setting.
Automated axe is not manual accessibility QA. Heading and landmark presence, focus order, screen-reader announcements, zoom/reflow, switch access, voice control, cognitive load, and real task completion still need a human matrix.
Static discovery excludes parameterized routes. Dynamic Studio states depend on their dedicated seeded specs and cannot inherit the full-static-tree claim.

Re-run evidence#

The current portable suites can be run against the real-development- infrastructure harness with one worker:

bash

PW_BROWSER_CHANNEL=chrome pnpm exec playwright test \
  -c apps/oshun/web/playwright.config.ts \
  apps/oshun/web/e2e/studio-route-render-smoke.spec.ts \
  apps/oshun/web/e2e/operator-surfaces.spec.ts \
  apps/oshun/web/e2e/operator-studio-surface.spec.ts \
  --workers=1

That command is intentionally large because the Studio spec collects every static route. A green run means the current render matrix, representative axe matrix, and named Operator flows passed; it still must not be reported as a manual AT sign-off or a zero-all-impact full-route axe sweep.

Source trail#

Cross-references#

Open questions#

Should CI add a scheduled all-479-route axe sweep while retaining the 39-route matrix as the fast per-change gate?
Which current Operator route/state matrix is required for release sign-off, including mobile, empty, error, permission-denied, and loading states?
Should the release gate fail moderate and minor axe findings, or report them through a separately owned accessibility debt budget?
Where will manual screen-reader, keyboard-only, 200–400% zoom, switch, and voice-control results be stored with browser/AT versions and task outcomes?