noir-test-flow

name	noir-test-flow
description	Visual + behavioral testing protocol for NOIR — triple-mode checks (light/dark/VI/responsive), interactive element testing, data consistency verification, and bug fix workflow. Used by /noir-qa orchestrator.

NOIR Test Flow — Visual + Behavioral Testing Protocol

Purpose: Defines HOW to test each page/feature. Called by /noir-qa and /noir-qa-run or used standalone. Standalone usage: /noir-test-flow — applies the full protocol to a specific page or set of pages. As sub-protocol: Referenced by /noir-qa Phase 4 (Execution) and /noir-qa-run Execute phase.

---

CORE PRINCIPLE

"If a human QA engineer would catch it, you must catch it."

Dark mode, Vietnamese, and responsive are NOT separate phases — they are tested on EVERY visual state (page, tab, dialog, popup). This catches the bugs that single-mode testing misses.

---

TRIPLE-MODE VISUAL AUDIT (Page Level)

For each page's initial view, execute 4 modes:

Light Mode (Baseline)

1. browser_navigate → page URL (or SPA sidebar click — see Technical Notes)
2. browser_wait_for → network idle / specific element visible
3. browser_take_screenshot → temp/qa-{page}-light.png
4. READ screenshot and analyze:
   - Layout: alignment, spacing, overflow, content clipping
   - Components: broken images, missing icons, wrong empty states
   - Typography: font consistency, text truncation, ellipsis working
   - Colors: theme consistency, contrast, hover states visible
   - Loading: skeleton → data (no flash of empty)
   - Card shadows: shadow-sm + hover:shadow-lg per design standard

Dark Mode

5. browser_evaluate → document.documentElement.classList.add('dark')
6. browser_take_screenshot → temp/qa-{page}-dark.png
7. READ screenshot and verify:
   - Text readable (no white-on-white, no black-on-black)
   - Borders visible (not disappearing into background)
   - Status badges maintain contrast
   - Charts/graphs readable
   - No hardcoded colors overriding theme variables
   - Input fields: background distinguishable from page background
   - Dropdown menus / popovers: correct dark background (not white)

Vietnamese

8. browser_evaluate → localStorage.setItem('i18nextLng','vi'); location.reload()
9. browser_wait_for → content loaded
10. browser_take_screenshot → temp/qa-{page}-vi.png
11. READ screenshot and verify:
    - No raw i18n keys (e.g., "products.searchPlaceholder" as literal text)
    - No English text remaining (except allowed: CRM, API, SMTP, Blog)
    - No text overflow from longer Vietnamese strings (buttons, headers, labels)
    - Date format matches Vietnamese locale (DD/MM/YYYY)
    - Sidebar labels are pure Vietnamese (no "Blog Posts", must be "Bài viết")
    - Vietnamese + dark mode: switch to dark briefly → screenshot → verify no issues
    - Switch back to light mode after

Responsive

12. Restore to light mode + English before responsive checks
13. For each viewport [1440, 1024, 768]:
    - browser_resize → {width}px
    - browser_take_screenshot → temp/qa-{page}-{width}px.png
    - READ and verify:
      - 1440: full layout, sidebar visible, all columns shown
      - 1024: sidebar may collapse, table columns may hide, no overflow
      - 768: mobile/tablet layout, content stacks vertically, no horizontal scroll on body
    - Specific: tables horizontal scroll within container, buttons not cut off,
      cards stack vertically, PageHeader wraps correctly, search remains usable
14. browser_resize → restore 1440px

---

MINI TRIPLE CHECK (Element Level)

For EACH interactive element type on the page. Every distinct visual state gets this protocol:

1. Screenshot in light mode (baseline)
2. Switch to dark mode → screenshot → verify: text readable, borders visible, inputs distinguishable
3. Switch to Vietnamese → screenshot → verify: all text translated, no overflow, no raw i18n keys
4. Resize to 768px → screenshot → verify: layout adapts, content accessible, no overflow
5. Restore light mode + English + 1440px before proceeding

Required Visual States (100% Matrix)

Visual State	Light	Dark	VI	768px	Notes
Each TAB content	✅	✅	✅	✅	Different content layout per tab
Create dialog	✅	✅	✅	✅	Form fields, buttons, footer
Edit dialog	✅	✅	✅	✅	May differ from create (pre-populated)
Detail/View dialog	✅	✅	✅	✅	Read-only content, labels
Confirmation dialog	✅	✅	✅	✅	Destructive button styling, text
Filter popover	✅	✅	✅	✅	Dropdown options, filter labels
Empty state	✅	✅	✅	✅	Icon, title, description text
Error banner (FormErrorBanner)	✅	✅	✅	✅	Banner background, error text
Validation errors on form	✅	✅	✅	✅	Error text color vs input bg
Toast/notification	✅	✅	✅	✅	Background contrast, text, position
Date picker popup	✅	✅	✅	✅	Month/day names, calendar grid bg
Dropdown menu (actions ⋮)	✅	✅	✅	✅	Menu item text, background, hover
Select/Combobox dropdown	✅	✅	✅	✅	Option labels, search, selected state
Color picker popup	✅	✅	✅

Efficiency: Shared components (DataTablePagination, etc.) need full Mini Triple Check only on FIRST occurrence. Subsequent pages: quick dark mode spot-check. Page-specific dialogs/tabs: always full check.

---

INTERACTIVE ELEMENT TESTING

Tables (DataTable)

• Column headers render correctly, all have labels
• Sort: click each sortable column header, verify order + indicator
• Pagination: page 2, change page size, verify "Showing X of Y" updates
• Search: type query → verify filtering → clear → verify reset
• Column visibility: open dropdown, hide column, verify gone, show back
• Column reorder: if supported, drag column, verify new order
• Density toggle: switch compact/normal/comfortable if available
• Empty state: search nonsense → verify <EmptyState> component (not plain text) → Mini Triple Check on empty state
• Row selection: check/uncheck rows, verify bulk action toolbar appears
• Actions dropdown: click ⋮ → verify menu items → verify each action works
• Group by: if supported, group → verify group headers with correct count

Forms (Create/Edit Dialogs)

• Open Create dialog → screenshot → Mini Triple Check
• Field-by-field testing for EVERY field:
  • Focus → verify focus ring/border
  • Tab to next → verify tab order (top-to-bottom, left-to-right)
  • Required fields → verify red asterisk * on label
  • Type invalid → blur → verify error (NOT while typing)
  • Type valid → verify error clears
  • Dropdowns: open → select → verify selection
  • Date pickers: open → select → verify format. Dark mode: calendar popup screenshot
  • Rich text editors: type → verify toolbar works
  • File uploads: upload → verify preview
  • Color pickers: select → verify preview
• Submit empty → verify ALL required field errors at once
• Submit invalid → verify specific field errors (not generic toast)
• Submit valid → verify: toast + dialog closes + list updates + data matches
• Cancel → closes, no data changed
• Click outside → closes (Credenza behavior)
• ESC → closes
• Edit mode: open existing → verify fields pre-populated → Mini Triple Check

Filters

• Apply each filter individually → verify results update
• Apply combination → verify AND logic
• Verify active filter indicator/badge shows count
• Clear individual → verify results update
• Clear all → verify full reset
• Filter persists across pagination

Destructive Actions

• Click delete/archive → verify confirmation dialog appears (NEVER instant delete)
• Mini Triple Check on confirmation dialog
• Cancel → nothing changed
• Confirm → item removed/updated + toast + related data updated + activity timeline

Tabs

• For EACH tab:
  1. Click tab → verify content loads
  2. Verify URL updates with ?tab=xxx
  3. Mini Triple Check on tab content
  4. Direct-navigate to ?tab=xxx → verify correct tab active + content loaded
  5. Verify no flash/flicker on tab switch

Messages & Notifications

• Toasts: trigger success action → verify position, auto-dismiss, text → Mini Triple Check
• Error banners: trigger server error → verify shows, dismiss works, clears on retry → Mini Triple Check
• Validation messages: submit invalid → verify under field → Mini Triple Check
• Date picker popup: open → verify calendar → Mini Triple Check
• Select/Combobox: open → verify options → Mini Triple Check
• Tooltip: hover icon button → verify shows → Dark + VI check (no resize needed)

Navigation

• Breadcrumbs: correct path, each segment clickable
• Sidebar: current page highlighted
• Back navigation: browser back works correctly
• Deep link: navigate to URL directly → verify same state loads

---

DATA CONSISTENCY CHECKS

After EACH CRUD operation, verify:

Create

• Item appears in list with correct data
• List total count incremented
• Dashboard widgets updated (if applicable)
• Activity timeline shows "Created ..." entry
• Related entities reflect change (e.g., department employee count)
• Search returns the new item

Edit

• ALL changed fields persisted (re-open edit dialog to verify)
• List row updated with new values
• Detail page (if exists) shows updated data
• Activity timeline shows "Updated ..." entry
• Related views reflect change

Delete (Soft)

• Item removed from active list
• List total count decremented
• Related entities updated (counts, references)
• Activity timeline shows "Deleted ..." entry
• Dashboard widgets updated

---

CROSS-FEATURE FLOW TEMPLATES

Use these as templates when creating flow definitions in .qa/flows/.

Flow: Product → Order → Payment Lifecycle

Create Product → Add Variants → Set Inventory (StockIn, confirm)
→ Dashboard product count increased
→ Create Customer → Create Order with product variant
→ Confirm → inventory decreased → Process → Ship → Deliver → Complete
→ Payment in Payments list with correct amount
→ Revenue in Dashboard
→ Customer writes Review → Approve → review count on product detail

Flow: Customer Journey

Create Customer → Add to Group → group count updated
→ Create Order → Customer Detail → order in history tab
→ Cancel order → status updates in: Order detail, Customer history, Dashboard
→ Activity Timeline shows all actions with correct descriptions
→ Vietnamese → activity descriptions translated

Flow: Content Publishing

Create Blog Category → Create Blog Tags (2+)
→ Create Post with image → Assign category + tags → Draft
→ Media Library shows image
→ Edit → Publish → status badge changes
→ Delete tag → post's tag list updates (no crash)
→ Edit category name → post shows updated category

Flow: HR Workflow

Create Department (child) → tree shows it
→ Create Employee in department → Add Tags
→ Department detail: employee count = 1
→ Edit employee → changes persist
→ Export CSV → CSV contains employee
→ Import modified CSV → change applied
→ Move to different department → both counts update

Flow: CRM Pipeline

Create Company → Create Contact linked → company shows contact count
→ Create Lead → Kanban shows in first stage
→ Drag through stages → Win → dashboard metrics update
→ Activity on contact → timeline shows
→ Delete company → contacts still accessible (soft delete)

Flow: PM Workflow

Create Project → auto-code PRJ-xxx
→ Add columns → Create tasks in different columns
→ Drag task → status changes in detail
→ Subtask → complete → parent progress
→ Comment → timestamp correct
→ Attachment → download works
→ Archive → gone from Kanban, in Archived view

Flow: Data Integrity & Edge Cases

Delete Brand with products → products still display
Disable Module (CRM) → sidebar gone, URL → 404, dashboard widgets hidden, re-enable → returns
Test user with custom role → remove permission → verify sidebar/URL/actions restricted
Language switch EN→VI→EN → all labels correct, no raw keys, date formats change
Theme switch light→dark→light → no artifacts

Flow: Error Handling

Navigate to /portal/entity/00000000-...0 → 404 or graceful redirect
Stop backend → Submit form → error banner (not crash) → Restart backend
Delete entity via API while edit dialog open → Save → conflict handling
Rapid pagination clicks → final page correct (no stale data)
Rapid double-click Submit → only ONE entity created

---

TEST CASE PRIORITY LEVELS

Priority	Meaning	When to run	Typical bugs found
P0	Critical path — app unusable if broken	Every commit (smoke)	CRITICAL, HIGH
P1	Core functionality — feature broken	Every PR (regression)	HIGH, MEDIUM
P2	Edge cases — non-obvious scenarios	Weekly / pre-release	MEDIUM, LOW
P3	Cosmetic — nice-to-have	Monthly / manual	LOW

BUG SEVERITY CLASSIFICATION

Severity	Definition	Examples
CRITICAL	Feature broken, data loss, security	Form submits but data not saved, crash, security bypass
HIGH	Feature partially broken, bad UX	Wrong data displayed, dialog won't close, pagination broken
MEDIUM	Visual bug, minor UX	Misalignment, wrong icon, dark mode contrast issue
LOW	Cosmetic, nice-to-have	Hover inconsistent, spacing off by a few px

Priority vs Severity: Priority is the test's urgency (when to run). Severity is the bug's impact (how bad). A P2 edge-case test can find a CRITICAL bug — that's the point of edge-case testing.

---

BUG FIX WORKFLOW

When you find a bug:

1. Log it — add to results with severity + description + screenshot
2. Analyze root cause — read relevant source code
3. Fix it — minimal code change
4. Restart if needed:
   • Frontend CSS/TSX/JSON → HMR, wait 2s
   • Backend handler/service → dotnet watch auto-reloads, wait 3-5s
   • DI/Program.cs → restart backend
   • vite.config → restart frontend
5. Build check (only if logic change):
   • Frontend: cd src/NOIR.Web/frontend && pnpm run build
   • Backend: dotnet build src/NOIR.sln
6. Health check:

   curl -sf http://localhost:4000/robots.txt -o /dev/null && echo "Backend: OK"
   curl -sf http://localhost:3000 -o /dev/null && echo "Frontend: OK"
   

7. Re-verify — exact same test that found the bug
8. Screenshot fixed state → temp/qa-{page}-fixed-{BUG-ID}.png
9. Update result status to FIXED+VERIFIED
10. Regression check — if shared component changed, note affected pages

Bug Decision Tree

CSS/layout bug? → Fix Tailwind classes → No restart (HMR)
Missing translation? → Add to BOTH en/*.json AND vi/*.json → No restart (HMR)
Dark mode bug? → Fix hardcoded color or missing dark: variant → No restart
Data display bug? → Trace component → hook → handler → spec → Fix at correct layer → Watch reload
Form validation? → Check Zod + FluentValidation → Fix both if needed → Watch reload
State management? → Check TanStack Query invalidation → No restart
Navigation/routing? → Check router, useUrlTab/useUrlDialog → No restart
DI/startup? → Fix Program.cs/DI → Restart backend
Field-level UX? → Fix FormField/FormItem props → No restart

FIX-THEN-COMMIT Strategy

After fixing ALL bugs on a single page (not per-bug), create ONE commit:

git add [specific files only]
git commit -m "$(cat <<'EOF'
fix(qa): [page-name] — [N] bugs fixed

- BUG-001: [brief description]
- BUG-002: [brief description]

QA: NOIR QA Agent

Co-Authored-By: Claude Opus 4.6 (1M context) <[e-mail]>
EOF
)"

---

TECHNICAL NOTES

Playwright MCP Tools

Use mcp__playwright__* for ALL browser interactions:

• browser_navigate — go to URLs (initial load only — see SPA bug below)
• browser_click — click elements (use accessible name or CSS selector)
• browser_fill_form — fill multiple form fields at once
• browser_take_screenshot — capture to temp/
• browser_snapshot — accessibility tree (for finding elements)
• browser_select_option — <select> dropdowns
• browser_press_key — keyboard (Escape, Enter, Tab)
• browser_wait_for — network idle or element visible
• browser_hover — hover states
• browser_drag — drag-and-drop (Kanban, tree reorder)
• browser_type — type character by character
• browser_resize — responsive breakpoints

Strategy: browser_snapshot first to understand page structure, then use accessible names for clicking/filling. More robust than CSS selectors.

CRITICAL: MCP Direct — No Scripts

NEVER write Playwright test scripts (.js/.ts files). Always use mcp__playwright__* tools directly with AI reasoning (max effort, Opus model). The AI reads screenshots, analyzes accessibility trees, and makes intelligent decisions about what to click/verify — this catches visual bugs that scripted tests miss.

❌ WRONG: Write a .js file → run it with npx playwright test
✅ RIGHT: mcp__playwright__browser_navigate → mcp__playwright__browser_snapshot → AI analyzes → mcp__playwright__browser_click

The entire value of this QA system is AI visual reasoning + interactive exploration, not scripted automation.

Windows SPA Navigation Bug (CRITICAL)

On Windows, browser_navigate / page.goto() often produces blank white pages for React SPA routes. HTML shell loads but React doesn't mount.

Workaround: Navigate via SPA sidebar clicks instead of direct URL. Use browser_navigate ONLY for initial page load (/portal), then sidebar clicks for all subsequent pages. For detail pages without sidebar links, click table rows from parent list.

Context Efficiency

• Save browser_snapshot output to files (filename param) instead of inline
• NEVER dump full sidebar accessibility tree — 100+ lines of identical content
• Only read <main> content area
• For DataTable pattern pages, screenshot is sufficient — skip tree dump

Login Flow

1. browser_navigate → http://localhost:3000/login
2. browser_snapshot → find email/password fields
3. browser_fill_form → email: [e-mail], password: 123qwe
4. browser_click → Sign In button
5. browser_wait_for → dashboard loaded
6. For Platform Admin tests: logout, login as [e-mail] / 123qwe

Screenshot Naming

temp/qa-{page}-{variant}.png

Page-level:  -light, -dark, -vi, -1024px, -768px
Element-level: -tab-{name}-{mode}, -create-dialog-{mode}, -edit-dialog-{mode}
               -confirm-dialog-{mode}, -empty-state-{mode}
Action:      -create-error, -create-valid, -fixed-{BUG-ID}

---

SERVICE LIFECYCLE

Restart Protocol

Change Type	Restart?	Action
Frontend CSS/TSX/i18n JSON	NO	Vite HMR (wait 2s)
Backend handler/spec/service	NO if dotnet watch	Watch reloads (wait 3-5s)
Backend DI / Program.cs	YES — backend	Kill :4000, rebuild, restart
vite.config / tailwind.config	YES — frontend	Kill :3000, restart pnpm dev

Restart Commands

# Restart backend
netstat -ano | grep ":4000 " | grep "LISTEN" | awk '{print $5}' | sort -u | while read pid; do taskkill //F //PID "$pid" 2>/dev/null; done
sleep 2
cd src/NOIR.Web && dotnet build --nologo -v q -c Debug > /dev/null 2>&1
ASPNETCORE_ENVIRONMENT=Development ASPNETCORE_URLS="http://localhost:4000" dotnet watch run --no-launch-profile > ../../.backend.log 2>&1 &

# Restart frontend
netstat -ano | grep ":3000 " | grep "LISTEN" | awk '{print $5}' | sort -u | while read pid; do taskkill //F //PID "$pid" 2>/dev/null; done
sleep 1
powershell -Command "Start-Process cmd -ArgumentList '/c cd /d src\NOIR.Web\frontend && pnpm run dev > ..\..\..\.frontend.log 2>&1'"

# Health check
curl -sf http://localhost:4000/robots.txt -o /dev/null && echo "Backend: OK" || echo "Backend: FAILED"
curl -sf http://localhost:3000 -o /dev/null && echo "Frontend: OK" || echo "Frontend: FAILED"

Crash Recovery

1. Check which service is down: tail -20 .backend.log or .frontend.log
2. If crash caused by your fix → revert fix first
3. Restart crashed service
4. Wait for health check
5. Resume testing