UX Waiting States Audit

Evaluate how applications handle long-running operations (30+ seconds) using browser automation.

Core Principle

Screenshot First, DOM Second — always take a screenshot when navigating or stuck. Visual inspection beats DOM probing.

📸 Screenshot → 👀 Analyze visually → 🎯 Click coordinates → 📸 Verify → Repeat

---

Critical: Screenshot-First Automation

When to Screenshot

ALWAYS screenshot:

• After opening any URL (initial state)
• Before trying to interact with elements
• When ANY JavaScript returns "missing value" or fails
• After clicking/submitting (verify success)
• At timed intervals during long operations

Why Screenshots Beat DOM Probing

DOM Approach	Screenshot Approach
Complex selectors fail silently	Visual shows exact UI state
"missing value" gives no info	Image reveals button locations
10+ attempts to find element	1 screenshot → click coordinates
Can't see actual user experience	See exactly what user sees

Screenshot Workflow Pattern

# 1. Navigate
Control Chrome:open_url(TARGET_URL)
sleep(2)

# 2. ALWAYS screenshot first
# Use browser screenshot or html2canvas
# Analyze the image before ANY interaction

# 3. If interaction needed, prefer coordinates over selectors
# After seeing screenshot: "The submit button is at ~(1200, 650)"
Control Chrome:execute_javascript("document.elementFromPoint(1200, 650).click()")

# 4. Screenshot again to verify

---

Quick Start Workflow

Phase 1: Setup & Navigate

# 1. Navigate to target URL
Control Chrome:open_url(TARGET_URL)
sleep(3)

# 2. SCREENSHOT - See what loaded
# Analyze: What's visible? Where are interactive elements?

# 3. Ask user to help identify:
#    - Which operation to test
#    - How to trigger it (button location, input needed)

Phase 2: Trigger Operation (The Hard Part)

This is often where audits get stuck. Modern SPAs have complex UIs.

# Strategy 1: Ask user for guidance
# "I see the page. Can you describe where the button is or what to click?"

# Strategy 2: Use simple, targeted JS
Control Chrome:execute_javascript("document.querySelector('button[type=submit]').click()")

# Strategy 3: Coordinate-based clicking (after screenshot)
Control Chrome:execute_javascript("document.elementFromPoint(X, Y).click()")

# Strategy 4: Let user trigger manually
# "Please click the button to start the operation, then tell me when it's processing"

Phase 3: Capture Waiting States

Once operation is running:

# T+0s: Screenshot immediately when operation starts
# T+10s: Screenshot after 10 seconds
sleep(10)
# Screenshot + capture_state.js

# T+30s: Screenshot after 30 seconds  
sleep(20)
# Screenshot + capture_state.js

# T+Complete: Screenshot when done
# Watch for UI changes indicating completion

Phase 4: Evaluate & Report

# 1. Evaluate screenshots against checklist (see references/checklist.md)
# 2. Generate report with annotated screenshots
# 3. Prioritize recommendations

---

Troubleshooting: When Stuck

Problem: Can't find/click element

❌ WRONG: Keep trying different selectors
   → Wastes time, silent failures

✅ RIGHT: Take screenshot, analyze visually
   → Ask user for help if needed
   → Use coordinate-based clicking

Problem: JavaScript returns "missing value"

This usually means:

1. The script is too complex (simplify it)
2. The element doesn't exist (screenshot to verify)
3. Timing issue (add sleep, retry)

Fix: Use simple one-liner JS, not complex functions.

// ❌ Complex (fails silently)
(function() { const elements = []; ... return JSON.stringify(elements); })()

// ✅ Simple (clear result)
document.body.innerText.substring(0, 500)
document.querySelectorAll('button').length
document.querySelector('.loading') !== null

Problem: Form won't submit

Try in order:

1. Screenshot to see submit button location
2. document.forms[0].submit()
3. Click submit button by coordinates
4. Ask user to submit manually

---

Audit Instructions

Step 1: Understand the Target

Ask user for:

1. URL to audit
2. Operation to test (search, report generation, AI task, etc.)
3. How to trigger — button name, location, or steps
4. Expected duration (helps calibrate screenshot intervals)
5. Any auth requirements (login credentials if needed)

If user is available: Ask them to trigger the operation manually while you capture screenshots. This avoids navigation complexity.

Step 2: Capture Sequence

Always screenshot first. Then run simple state checks:

// Simple one-liners that won't fail silently:

// Check for spinner
!!document.querySelector('[class*="spin"], [class*="load"], .spinner')

// Check for progress bar
!!document.querySelector('progress, [role="progressbar"]')

// Get visible text (look for status messages)
document.body.innerText.substring(0, 1000)

// Count results appearing
document.querySelectorAll('[class*="result"], [class*="item"]').length

Capture Timeline: | Time | Action | |------|--------| | T+0s | Screenshot + note what triggered | | T+10s | Screenshot + simple state checks | | T+30s | Screenshot + simple state checks | | T+Complete | Screenshot + final state |

Step 3: Evaluate Against Checklist

Load and evaluate against references/checklist.md. Score each category:

• ✅ Present and well-implemented
• ⚠️ Partially implemented / needs improvement
• ❌ Missing
• N/A Not applicable to this operation

Step 4: Generate Report

Use template from references/report-template.md.

---

Key Evaluation Patterns

Progressive Value Detection

Look for:

// Partial results appearing
document.querySelectorAll('[class*="result"], [class*="item"], li, tr').length

// Streaming content
document.querySelector('[class*="stream"], [class*="typing"], [class*="cursor"]')

Heartbeat Indicators

Look for:

// Counters
document.body.innerText.match(/\d+\s*(found|processed|complete|%)/gi)

// Animations (CSS or JS)
document.querySelectorAll('[class*="animate"], [class*="pulse"], [class*="spin"]')

Time Estimation

Look for:

// Time remaining text
document.body.innerText.match(/(\d+\s*(sec|min|second|minute)|remaining|left|ETA)/gi)

// Progress percentage
document.querySelector('[role="progressbar"]')?.getAttribute('aria-valuenow')

---

Screenshot Comparison Strategy

For each interval, note:

1. What changed from previous screenshot
2. Activity signals (counters, animations, partial results)
3. User anxiety factors (frozen UI, no feedback)

Compare: | Element | T+0s | T+10s | T+30s | Complete | |---------|------|-------|-------|----------| | Results visible | | | | | | Counter/progress | | | | | | Status message | | | | | | Animation active | | | | |

---

Report Output Structure

Generate markdown report with:

1. Summary Score: X/10 categories addressed
2. Screenshots Gallery: With timestamps and annotations
3. Strengths: What works well
4. Critical Gaps: Missing elements hurting UX most
5. Quick Wins: Low-effort, high-impact improvements
6. Detailed Findings Table: See references/report-template.md
7. Priority Matrix: P1/P2/P3 recommendations

---

Best-in-Class Comparisons

Reference these examples of excellent waiting UX:

• Figma exports: Progress bar + percentage + file count
• Notion AI: Streaming text + cursor animation
• ChatGPT: Token-by-token streaming + stop button
• Linear search: Instant partial results + refinement
• Vercel deployments: Step-by-step progress + logs

---

Error Scenarios to Test

If possible, test:

1. Partial failure: Does UI degrade gracefully?
2. Network interruption: Is progress preserved?
3. Timeout: Is there clear feedback?

// Simulate slow network (if DevTools available)
// Or disconnect briefly and observe behavior

---

Common Issues to Flag

Issue	User Impact	Quick Fix
Spinner only	Anxiety, abandon	Add status text
No progress	"Is it stuck?"	Add heartbeat counter
No cancellation	Trapped feeling	Add cancel button
Silent completion	Missed results	Add completion animation
Full-page block	Can't multitask	Move to background