cesnimda
14 KiB
Manual acceptance cards — P9-REQUAL-01
All M09-01–10 remain NOT TESTED by a human in this requalification. These cards expand the unchanged original checklist; they record no fabricated execution. Under REQUEST §6, unfinished checks carry into release gates. They are not automatically57 preservation defects or proof of a present failure. Relevant RD06/07/10/14/18/23/25 qualifications remain independent of local automation.
Before each card, record exact final application commit, static-export/API assembly/source-manifest hashes, environment URL, MariaDB image/config digest and effective nonproduction configuration. Use the final qualified build from the root validation ledger; no card may silently run against a stale build. Record tester name/role, UTC start/end, OS/browser/AT/device/font/printer versions and test-data IDs. Use isolated development or specifically approved sandbox environments, never production. If any environment/permission/device is absent, leave that applicable portion NOT TESTED and record the dependency; no implication of failure or pass.
Synthetic pack: one owner plus separate admin and foreign tenant accounts, two independent same-owner sessions;250 jobs only when the approved isolated fixture provides them; one ordinary job with long title/notes/timeline; long Career history with8 roles/48 distinct bullets, Åse/æ/ø, one-/multi-page CVs, optional sections and synthetic photo; notifications0/1/many, all9 Dashboard widgets. Use controlled provider inboxes for M09-10. Verify seed readback first. Do not create/delete/reset live records. Preserve frozen document hashes before/after any relevant workflow. Controlled failure injection is an approved harness action, not an invitation to stop a real service.
Priority P1 identifies release/accessibility/data/output/provider qualification; P2 is professional language review. Priority does not grant production or provider authority. Complete cards against both English and Bokmål and light/dark where applicable, retaining per-state evidence rather than one unexplained aggregate checkmark.
M09-01 — Reader landmarks and complete forms (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: NVDA with a supported Windows browser, then VoiceOver with Safari on supported Apple hardware. Record actual versions.
Steps: With reader speech enabled, traverse /login, /jobs, /jobs/{synthetic-id}, /career, /career/builder/{synthetic-id}, /documents and /settings by headings, landmarks, tab/form navigation. Trigger an invalid field and a failed save in the approved test harness, then recover. Open and settle-close drawer/dialog/menu. Repeat equivalent journey with VoiceOver.
Pass criterion: Names/roles/values/selection, hierarchy, errors and reading order understandable; modal background not announced; successful/error status accurately announced.
Fail criterion: Missing or misleading name/state, hidden content spoken, unreachable form or unannounced blocking error.
Evidence to retain: Spoken transcript or recording, reader settings and exact failing/successful control; no real personal data.
M09-02 — Focus and modal transitions (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: Supported desktop browser, keyboard only; repeat drawer at narrow width. Actual reader can supplement, never substitute keyboard navigation.
Steps: Start at skip link; Tab/Shift+Tab through navigation and command dialog. Open Add/Edit Job, confirmation, select and menu; Cancel, Escape and submit/retry after a controlled failed save. Move through job sections and CV preview/history. Open/close drawer repeatedly including during ordinary transitions.
Pass criterion: Visible logical focus; modal trap only while open; Escape/cancel returns focus to initiating usable control; no closed/offscreen tab stops or lost focus.
Fail criterion: Keyboard trap outside modal, disappearing focus, background tabbing during modal or inability to recover after failure.
Evidence to retain: Short video plus focus sequence/control names and transition timing; preserve historical RD14 observation even if current run passes.
M09-03 — Dashboard announcements and deliberate recovery (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: Supported browser and actual reader; two independent authenticated sessions for same synthetic owner. Approved isolated harness for failed response.
Steps: On /dashboard open Customize. Hide/show a widget, Earlier/Later, resize, Cancel; reopen, Reset cancel then confirm. Create stale layout in second session and save it; first session Save must conflict, retain draft and offer review/reload. Inject one save failure using approved harness, then deliberately recover. Save empty layout and restore visible widgets.
Pass criterion: Each action/order/save/error announced once, spoken position agrees with DOM order, keyboard alternative to pointer exists, conflicts/failures retain draft and never claim saved.
Fail criterion: Silent/misleading status, inaccessible reordering or lost rejected draft.
Evidence to retain: Reader transcript, before/after order/revision and response status; do not capture credentials.
M09-04 — Contextual painted contrast (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: Actual supported browser, light/dark, painted-color sampling tool; include focused/hover/disabled/native-select/scrolled states.
Steps: Use original P9-AX-02/03 affected nodes from inherited triage. Inspect size controls, all53 painted-background occurrences grouped by cause, scrolled CV editor and fixed chrome. Open native select and inspect applicable text/UI/focus in both themes; retain each affected cause/node/state mapping.
Pass criterion: Normal text >=4.5:1; large text >=3:1; necessary UI boundaries/focus >=3:1. Apply documented WCAG exceptions only to truly inactive controls; record why.
Fail criterion: Any applicable ratio below threshold or no reliable painted sample for a required state; unresolved sample stays NOT TESTED.
Evidence to retain: Foreground/background actual paint colors, ratio, node/state and screenshot; automation incomplete alone is not closure.
M09-05 — Hidden badges and settled drawer (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: Actual reader plus browser accessibility tree; Dashboard and original Analytics drawer context.
Steps: Inspect unread=0 hidden badge, then synthetic positive unread count. Open drawer; inspect modal background and reading/tab order; settle-close and repeat reader traversal and24 Tab presses. Repeat in Analytics context matching inherited occurrence.
Pass criterion: Zero badge absent from announcements, positive count meaningful, open modal excludes background appropriately; settled closed drawer neither spoken nor tabbable.
Fail criterion: Hidden zero announcement, missing positive count, closed drawer reader/focus access or background leakage during modal.
Evidence to retain: Accessibility tree, speech and focus record per exact state. Passing settled Dashboard does not retroactively explain original Analytics cause.
M09-06 — Reflow, clipping and visible controls (P1)
Owner: Accessibility tester. Status: NOT TESTED.
Environment: Real browser100/200/400% zoom, text-only enlargement where supported, increased text spacing and forced colors; record OS/browser and actual zoom method.
Steps: At each setting traverse Jobs pagination, job/document section tabs, CV tabs/controls, Settings inputs and significant forms using keyboard. Scroll intentional table/document/Board scrollers, then reach every primary action. Check long EN/NB labels and error states.
Pass criterion: Essential content and focused actions readable/reachable, no root clipping or lost focus; intentional scrollers remain usable; fixed A4/Letter document geometry preserved.
Fail criterion: Unreachable clipped control/text, obscured primary action or focus, broken forced-colors affordance.
Evidence to retain: Screenshot at each setting, focused control and scroll path. The historical16 offviewport states remain explicit targets; zero root overflow alone is insufficient.
M09-07 — Physical input and interruption (P1)
Owner: Product owner with physical devices. Status: NOT TESTED.
Environment: Actual supported phone and tablet, portrait/landscape, software keyboard, touch/pointer as applicable; record model/OS/browser/input.
Steps: Edit long Career/job/letter text while keyboard is open. Open drawer/dialog/select, rotate, recover after Cancel and controlled failed save. Reorder Dashboard/Board using supported touch action and named alternative; cancel drag. Traverse CV preview; download synthetic PDF.
Pass criterion: Primary action not obscured by keyboard; draft retained across intended interruptions; reachable touch targets; no unintended destructive/move operation; alternatives usable.
Fail criterion: Obscured save, lost draft, accidental destructive/move, unusable touch target or failed supported download.
Evidence to retain: Device video/screenshots including keyboard/rotation and result readback. Emulated touch cannot close this card.
M09-08 — Professional Bokmål and terminology (P2)
Owner: Professional Bokmål reviewer. Status: NOT TESTED.
Environment: Application nb locale on exact build; both themes, narrow and desktop, ordinary user and separate synthetic admin. Include dates/numbers/plural states.
Steps: Review registration/recovery, errors/validation, onboarding, Dashboard customization, Jobs/Board/history, Career/import, CV/library, documents/mail, notifications, Settings/Admin. Use long names,0/1/many counts and boundary date/number values. Inspect accessible names with reader. Record raw Created metadata separately from application copy.
Pass criterion: Natural professional Bokmål, consistent established terms, meaningful validation/names and correct date/number presentation. Distinguish authored/provider English and technical names.
Fail criterion: Untranslated application string, unnatural/misleading term, incorrect locale format or inaccessible label; authored English alone is not defect.
Evidence to retain: Exact key/control and screenshot, proposed wording and reviewer rationale. P9-L02 remains low debt until separately implemented/reviewed.
M09-09 — Intended-environment PDF fidelity (P1)
Owner: Document reviewer. Status: NOT TESTED.
Environment: Intended Linux renderer/font set, supported primary and alternate browsers and actual printer or approved print workflow. Record versions/fonts/template/page size.
Steps: Generate synthetic one- and multi-page EN/NB CVs in every supported template, with long words/employment/bullets, missing optional sections, Å/æ/ø/special characters and permitted synthetic photo. Compare preview with downloaded originals; inspect/search/copy text; print representative pages including first/last/page-break boundaries.
Pass criterion: Legible fonts/glyphs, correct language/date, no missing/duplicate content, intact margins/header/footer/page breaks, photo/crop and preview/export agreement.
Fail criterion: Missing glyph/content, overflow/cutoff, duplicate blocks, wrong pagination/font/photo or illegible printed output.
Evidence to retain: Original PDF bytes+SHA256, page count, renderer/font/browser versions, screenshots/print observations and named template. Existing60-page automated oracle is separate.
M09-10 — Real provider consent, delivery and uncertain outcomes (P1)
Owner: Approved provider tester. Status: NOT TESTED.
Environment: Explicitly authorized nonproduction OAuth/Turnstile/mail/calendar sandbox accounts, controlled recipient inboxes and timezone/DST scenarios. No production credentials or real personal documents.
Steps: Test Google/Microsoft consent and denied/cancelled popup; validate expected origin/state. Verify/reset email to sandbox inbox including used/expired token. Link mailbox, revoke grant and reconnect. Simulate uncertain send only with approved sandbox fixture, then review rather than automatically resend. Create calendar events around DST transitions, explicit offsets and date-only boundaries.
Pass criterion: Correct owner/origin/state, explicit confirmation, recoverable revoke/reconnect, actual intended mail delivery, no accidental resend and correct timezone/date semantics.
Fail criterion: Cross-owner/mismatched state accepted, resend without deliberate review, failed intended delivery or wrong event time/date.
Evidence to retain: Correlated sanitized provider request/response/delivery/event timestamps, exact account scope and timezone expectation. Local SMTP/HTTP tests do not close provider semantics.
Recording and closure
For each state, retain a row: card ID | tester | UTC | commit/build hashes | environment | OS/browser/AT/device | synthetic IDs | steps actually executed | expected | observed | PASS/FAIL/NOT TESTED | evidence path/hash | defect ID/owner/retest. A failed observation gets a bounded reproducible defect and linked evidence; keep the original failure, then record a new successor after repair. Untested subcases stay explicit; completing one browser/device never closes another. Redact tokens, credentials and unrelated personal information. Store recordings/reader transcript/PDF bytes in approved nonproduction evidence storage and link them from the final ledger.
A genuine tester reviews every applicable subcase before closing the card; product/accessibility/provider owners accept any documented non-applicable state with a reason. An AI screenshot review, DOM assertion, healthy endpoint or axe0-violation result cannot sign these cards. C11/P5-13 semantics, infrastructure/backup/restore/rollback, monitoring, creation idempotency and enabled-worker recovery have separate gates and are not authorized by this checklist. No Phase10 implementation or release approval follows automatically.