coalesce-labs/prototype-feedback

Prototype Feedback Loop Workflow

Rapidly iterate on prototypes using AI-powered building and automated feedback collection.

Quick Start

1. Share a prototype link (v0/Lovable/Bolt) or describe what you built
2. I check the related PRD, design system, and user research for context
3. I run a structured feedback analysis (PRD alignment, usability, multi-perspective)
4. I deliver prioritized recommendations: must-fix, should-fix, and nice-to-have
5. You iterate on the prototype and we repeat until validated

Example: "Review my checkout prototype: [link]. PRD is in thoughts/shared/pm/prds/checkout-redesign.md"

Output: Saved to thoughts/shared/product/prototypes/[feature]-feedback-round-[N].md

Time: 30 minutes per feedback round

Context Routing

Check these files before providing feedback:

| Source | Files/Folders | What to Extract | | -------------------- | --------------------------------------------------- | -------------------------------------------------- | | PRD | thoughts/shared/pm/prds/ | Requirements, acceptance criteria, success metrics | | Design System | design docs | Colors, typography, component patterns to match | | Stakeholder Profiles | thoughts/shared/pm/context/stakeholder-template.md | Who reviews this, their priorities and concerns | | User Research | thoughts/shared/pm/ | User pain points, quotes, behavior patterns | | Past Prototypes | thoughts/shared/product/prototypes/ | Previous feedback rounds, resolved issues |

Overview

Tools: v0/Lovable/Bolt + NotebookLM + Claude When: Validating new features before full build

---

Feedback Analysis Template

Use this structure for every feedback round:

PRD Alignment Check

| PRD Requirement | Prototype Status | Gap? | | --------------- | ------------------------------- | -------------------- | | [Requirement 1] | Implemented / Partial / Missing | [Description if gap] | | [Requirement 2] | Implemented / Partial / Missing | [Description if gap] | | [Requirement 3] | Implemented / Partial / Missing | [Description if gap] |

Usability Assessment

| Heuristic | Rating (1-5) | Issues | Recommendation | | ----------------------------------- | ------------ | ------ | -------------- | | Visibility of system status | | | | | Match between system and real world | | | | | User control and freedom | | | | | Consistency and standards | | | | | Error prevention | | | | | Recognition over recall | | | | | Flexibility and efficiency | | | | | Aesthetic and minimal design | | | |

Multi-Perspective Feedback

Engineering view: [Technical feasibility, performance concerns, implementation complexity, tech debt risks]

Design view: [Visual consistency, interaction patterns, accessibility gaps, design system alignment]

User view: [Ease of use, value clarity, friction points, learning curve, "would they actually use this?"]

Tool-Specific Feedback Adjustments

Adjust feedback focus based on which tool generated the prototype:

v0.dev prototypes (single components):

• Focus on: Component behavior, interaction states, visual polish, responsive behavior
• De-emphasize: Navigation flow, multi-page consistency, backend integration (v0 is component-level)
• Ask: "Does this component work in isolation? How will it integrate with the existing UI?"

Lovable/Bolt.new prototypes (full-stack apps):

• Focus on: End-to-end user flow, page-to-page navigation, data persistence, error handling
• De-emphasize: Pixel-perfect styling (these tools prioritize function over form)
• Ask: "Does the full flow work? Are there dead ends or missing states?"

Claude Artifacts (quick mockups):

• Focus on: Conceptual accuracy, layout structure, content hierarchy
• De-emphasize: Visual fidelity, interaction details (Artifacts are low-fidelity by nature)
• Ask: "Does this capture the right concept? Is the information architecture correct?"

design prototypes (designer-created):

• Focus on: Design system compliance, accessibility, edge case handling, micro-interactions
• De-emphasize: Technical feasibility (that's the engineer's feedback domain)
• Ask: "Is this usable? Does it handle real-world scenarios beyond the happy path?"

ASCII/napkin sketches (from /napkin-sketch):

• Focus on: Layout logic, information hierarchy, flow completeness
• De-emphasize: Everything visual (it's ASCII art, not a design comp)
• Ask: "Is the structure right? Are we missing any screens or states?"

Prioritized Recommendations

| # | Issue | Severity | Fix | Iteration | | --- | ---------------------- | -------- | -------------- | --------- | | 1 | [Must-fix before beta] | Critical | [Specific fix] | Current | | 2 | [Should-fix before GA] | Medium | [Specific fix] | Next | | 3 | [Nice-to-have polish] | Low | [Specific fix] | Later |

---

Workflow

Step 1: Build Initial Prototype

Using v0 (for UI components):

# Go to v0.dev
# Paste your feature description:

"Build a task management interface with:
- List view of tasks with checkboxes
- Ability to add new tasks
- Filter by status (all/active/completed)
- Clean, modern design similar to Linear

Include:
- Search functionality
- Due date display
- Priority labels (high/medium/low)"

v0 generates:

• React component
• Fully functional prototype
• Copy-paste ready code

Using Lovable (for full-stack apps):

# Go to lovable.dev
# Describe full feature:

"Build a customer feedback submission portal:
- Public form for submitting feedback
- Backend to store submissions
- Admin dashboard to review
- Email notifications on new submissions"

Lovable generates:

• Frontend + backend
• Database schema
• Working prototype with real functionality

Using Bolt (middle ground):

• Good for interactive prototypes
• Handles forms, validation, multi-step flows
• Between v0 (UI only) and Lovable (full stack)

Pro tip: Start with PRD from your PRD workflow, paste into AI builder, get 80% done automatically.

Step 2: Refine Prototype (30-60 min)

Iterate with AI:

"Make these changes:
1. Move the search bar to the top right
2. Add bulk actions (select multiple, mark complete)
3. Make priority labels more prominent
4. Add keyboard shortcuts (enter to add task, / for search)"

AI updates code in real-time. Test changes immediately.

Polish details:

• [ ] Error states (what happens when something fails?)
• [ ] Empty states (what shows when no data?)
• [ ] Loading states (what shows while loading?)
• [ ] Mobile responsive (does it work on phone?)

Step 3: Collect User Feedback (1 day)

Deploy prototype:

• v0: Export to CodeSandbox or Netlify
• Lovable: One-click deploy
• Bolt: Deploy to their hosting

Share with users:

Email template:

Subject: Quick feedback needed on [Feature] prototype

Hi [Name],

We're exploring [feature] and would love 15 minutes of your time to get feedback on a prototype.

Try it here: [link]
Then book time: [calendly link]

Or just reply with your thoughts!

Thanks,
[You]

Interview script:

1. Don't explain anything. Just share link.
2. Watch them try to use it (screen share).
3. Ask: "What do you think this does?"
4. Ask: "Try to [accomplish task]. Think out loud."
5. Note: Where do they get confused? What surprises them?
6. Ask: "Would you use this? Why/why not?"
7. Ask: "What's missing? What would you change?"

Capture everything:

• Record sessions (with permission)
• Take notes
• Screenshots of confused moments

Step 4: Synthesize Feedback (30 min)

Use NotebookLM:

1. Upload all interview transcripts
2. Upload session recordings (transcribe first)
3. Upload notes

Query:

"Analyze these user feedback sessions:

1. What patterns did you see?
   - Where did users get confused?
   - What worked well?
   - What didn't work at all?

2. Group feedback into themes

3. Rate each issue by:
   - Frequency (how many users hit this?)
   - Severity (how bad is it?)

4. Recommend top 3 changes for next iteration"

Output: Prioritized list of changes to make.

Step 5: Iterate Prototype (1-2 hours)

Make top changes:

# Back to v0/Lovable/Bolt

"Based on user feedback, make these changes:

High priority:
1. [Change 1 - 80% of users confused by X]
2. [Change 2 - 60% of users couldn't find Y]
3. [Change 3 - 100% of users asked for Z]

Update the prototype to address these issues."

AI updates the prototype. Test yourself.

Step 6: Second Round of Feedback (1 day)

Same process, new users (if possible):

• Show updated prototype
• See if issues are fixed
• Discover new issues

Or test with same users:

• "Here's the updated version"
• "Did we address your concerns?"
• "What else needs work?"

Step 7: Decide to Build or Iterate (30 min)

Success criteria checklist:

• [ ] Users understand what it does (>80%)
• [ ] Users can complete core tasks (>70%)
• [ ] Users would actually use this (>60%)
• [ ] No critical usability issues
• [ ] Feedback is mostly positive

If yes: Write real PRD and hand off to eng/design
If no: Another iteration or pivot

---

Automation Opportunities

Automated Feedback Collection

Using Typeform + Make.com:

1. User tries prototype
2. Typeform survey pops up automatically
3. Responses go to Airtable
4. Claude synthesizes daily
5. Slack notification with insights

Questions to ask:

• What were you trying to do?
• Did you accomplish it? (Yes/No)
• What was confusing?
• What would you change?
• Would you use this? (1-5 scale)

Automated Session Recording

Using FullStory or Hotjar:

• Tracks all user sessions
• Shows where users click, scroll, rage-click
• Heatmaps show attention patterns
• AI can analyze patterns

Continuous Feedback Loop

Weekly cycle:

1. Monday: Build/update prototype
2. Tuesday-Thursday: Collect feedback
3. Friday: Synthesize and iterate
4. Repeat until validated

---

Example: Full Cycle

Week 1:

• Built initial prototype (2 hours)
• Tested with 5 users (1 day)
• Found: Nobody understood the navigation
• Iterated: Simplified nav (1 hour)

Week 2:

• Tested new version with 5 users (1 day)
• Found: Nav fixed! But bulk actions confusing
• Iterated: Redesigned bulk actions (1 hour)

Week 3:

• Tested again with 3 users (1 day)
• Found: 100% success rate on core tasks
• Decision: Validated! Write real PRD

Total time: 3 weeks, ~12 hours of work
Alternative: 7-12 weeks, full build, then discover it's wrong

ROI: 6-9 weeks saved, $50K+ in eng time saved

---

Common Mistakes

Don't:

• ❌ Build production code (it's a prototype!)
• ❌ Make it perfect (good enough to test)
• ❌ Test with only internal team (get real users)
• ❌ Ignore negative feedback (embrace it)
• ❌ Iterate forever (set a limit: 3-4 rounds max)

Do:

• ✅ Build fast, test fast, learn fast
• ✅ Embrace rough edges (it's fine for a prototype)
• ✅ Get in front of real users ASAP
• ✅ Track feedback systematically
• ✅ Know when to kill bad ideas early

---

Tool Recommendations

For UI prototypes: v0.dev
For full-stack apps: Lovable.dev
For interactive flows: Bolt.new
For feedback synthesis: NotebookLM
For session recording: FullStory or Hotjar
For surveys: Typeform or Tally

Start free, upgrade if you're using daily.

---

Advanced: Multi-Variant Testing

Test 2-3 approaches at once:

Variant A: Approach 1 (e.g., wizard-style onboarding)
Variant B: Approach 2 (e.g., dashboard-style onboarding)
Variant C: Approach 3 (e.g., tutorial-style onboarding)

Split users:

• 5 users per variant
• See which performs best
• Build the winner

Faster learning, better outcomes.

---

Measuring Success

Prototype velocity:

• Ideas → prototype: <2 hours
• Prototype → feedback: <2 days
• Iterations per week: 2-3

Validation quality:

• User success rate: >70%
• Would-use rate: >60%
• Critical issues found: 0

Business impact:

• Features validated before build: 90%
• Failed features caught early: 100%
• Eng time saved: $50K+ per year

---

Time saved: 6-9 weeks per feature Cost saved: $50K+ per year in eng time Success rate: 90% (vs. 40% building without validation)

---

Output Quality Self-Check

Before delivering prototype feedback, verify:

• [ ] PRD alignment checked -- Every PRD requirement is mapped to prototype status (implemented, partial, missing)
• [ ] Usability heuristics scored -- At least 5 Nielsen heuristics rated with specific issues noted
• [ ] Multi-perspective feedback included -- Engineering, design, and user viewpoints are all represented
• [ ] Recommendations are prioritized -- Issues ranked by severity with clear fix descriptions and iteration assignment
• [ ] Context was checked -- PRD, user research, and stakeholder profiles were referenced (not just generic feedback)
• [ ] Actionable next steps -- PM knows exactly what to change in the next iteration
• [ ] Success criteria referenced -- Feedback connects back to PRD success metrics and kill criteria
• [ ] Previous feedback rounds referenced -- If this is round 2+, confirm previous issues were resolved

If any check fails, fix it before delivering. Generic feedback wastes iteration cycles.