Validate risky game mechanics or concepts quickly with throwaway code.
1-3 day sprints focused on answering exactly ONE question. Disciplined exit
criteria prevent scope creep and sunk-cost fallacy.
🤖 Multi-Platform Notes
Platform
Notes
OpenClaw
Can create prototype plan, track progress, and write results. Best for structured prototyping docs.
Claude Code
Can write prototype code directly. Best for actual implementation support.
Cursor/Windsurf
Excellent for coding the prototype. Can scaffold a minimal project quickly.
GitHub Copilot
Can help write prototype code inline. Great for quick implementations.
Generic LLM Chat
Guide the user through the process and provide code snippets.
Any AI
All can do the planning phase. Coding-capable AIs can help implement.
Key principle: A throwaway prototype should be THROWN AWAY. Do not let
prototype code become production code unless you're doing a game jam.
📋 Phase 1: Hypothesis Definition
Step 1.1 — The One Question
Ask:
"What is the ONE thing we need to validate?"
Good questions:
"Does this control scheme feel good?"
"Is this core loop fun for more than 5 minutes?"
"Can we render 1000 entities at 60fps?"
"Does the procedural generation produce interesting levels?"
Bad questions (too broad):
"Is the game fun?" (too vague)
"Will people buy it?" (can't test in 3 days)
Step 1.2 — Hypothesis Statement
Write as:
"We believe that [mechanic/concept] will [outcome] for [player type]."
Example: "We believe that magnetic polarity switching will create satisfying
traversal puzzles for puzzle-platformer fans."
Step 1.3 — Success Criteria (Exit Criteria)
Define measurable criteria:
markdown
## Success Criteria (3 = pass, ≥2 = consider, <2 = fail)
| # | Criterion | Measure |
|---|-----------|---------|
| 1 | Player can complete the loop without instructions | ≥ 3 of 3 playtesters |
| 2 | Average time to first "fun" moment | < 30 seconds |
| 3 | Players attempt to repeat the action unprompted | Observed in ≥ 2 testers |
| 4 | Framerate stays above 60fps | With target entity count |
| 5 | [Custom criterion] | [Measure] |
🎯 Phase 2: Scope
Step 2.1 — The Minimum Viable Prototype
Ask:
"What is the smallest thing we can build to test our hypothesis?"
Examples:
For "does this control feel good?" → one scene, one object, the control
For "is this loop fun?" → the loop with placeholder art, no menu
For "performance?" → stress test scene, no gameplay
Step 2.2 — What We WON'T Build
List explicitly:
text
WILL NOT:
- Menu screens
- Sound effects
- Tutorial
- Save/load
- Polish
- Art (placeholder only)
- UI
- Analytics
Step 2.3 — Constraints
Dimension
Limit
Time
1-3 days (fixed, non-negotiable)
Engine
Whatever gets it done fastest
Art
Primitives / placeholders / free assets
Code quality
"Working" not "clean" — it will be thrown away
🛠️ Phase 3: Build Sprint
Day 1: Core
Set up minimal project
Implement the ONE mechanic
No polish, no extra features
End of day: it should run and be testable
Day 2: Playability
Add simple inputs
Add a goal/objective so there's a "loop"
Add minimal feedback (visual indicator of success)
End of day: someone who isn't you can play it
Day 3: Playtest & Decision
Don't add new features
Polish only what's needed to remove blockers
Run playtest (see Phase 4)
Make decision (see Phase 5)
Daily Check-Ins
Each day, ask:
"What did you build today? Did anything surprise you? Are we still on track?"
If something surprising (good or bad) changes the hypothesis, adjust.
🎮 Phase 4: Playtest
Step 4.1 — Tester Selection
Ask:
"Who will test?"
text
1) Yourself (better than nothing, biased)
2) Friend/family (honest but kind)
3) Fellow dev (technically knowledgeable)
4) Target audience member (ideal)
Step 4.2 — Test Protocol
Provide testers the prototype and observe WITHOUT giving instructions.
Observe:
Do they figure it out without help?
Where do they get stuck?
Do they smile / curse / lean in?
Do they want to keep playing after the test?
What do they try that you didn't expect?
Ask after:
"What was confusing?"
"What was the moment it 'clicked'?"
"What would make this more fun?"
"What's one thing you'd change?"
Do NOT ask:
"Did you like it?" (they'll be polite)
"Is this a good idea?" (they don't know)
Step 4.3 — Score Against Criteria
Rate 1-5 for each criterion. Collect data below.
🚦 Phase 5: Decision
Step 5.1 — The Three Doors
Outcome
Criteria
Action
PROCEED 🟢
Most criteria met; data supports hypothesis
Move to production. Archive the prototype. Start fresh with real code.
PIVOT 🟡
Mixed results; interesting but needs change
Document what didn't work. Modify hypothesis. Run another prototype sprint (1 day max).
KILL 🔴
Hypothesis disproven or not worth pursuing
Celebrate the learning. Document findings. Move to next concept. NO sunk-cost.
