
test-thinking
by dhmilnes
Claude code for product
SKILL.md
name: test-thinking description: Lightweight reference for designing meaningful test variations.
A/B Test Mindset
Lightweight reference for designing meaningful test variations.
The goal in testing is to build strong variations that create signal. We don't necessarily want to isolate a single variable — we want to create distinct experiences that have a chance in breaking through the noise.
Variation Design Framework
Step 1: Identify the Mechanisms at our disposal.
Is the change noticable to users? (above the fold, in the first 5 seconds, different flow, etc.)
Are there usability insights we have discovered?
Is there a psychological lever we are pulling to increase motivation? Name it.
Step 2: Design Distinct Variants
Bad: Control vs slightly different version Good: Control vs fundamentally different approaches to the same goal
Example (Progress Bar):
- Control: No progress indicator
- Variant B: Explicit steps ("Step 2 of 5") — tests clarity/goal gradient
- Variant C: Visual bar only (no numbers) — tests progress without anchoring
Step 3: Validate Each Variant
For each variant, answer:
- What evidence supports this approach?
- Why might this win?
- Why might this lose?
If you can't answer #1, it's not a real test — it's a guess.
Behavioral Science Quick Reference
Motivation & Goals
| Principle | Definition | Example |
|---|---|---|
| Goal Gradient | Effort increases as goal approaches | Progress bar, "You're 80% there!" |
| Endowed Progress | People value progress they "received" | Start bar at 20%, show completed steps |
| Sunk Cost | Past investment influences future decisions | "You've already completed 2 steps" |
| Fresh Start Effect | New beginnings motivate change | "Start your year strong" |
Prospect Theory
| Principle | Definition | Example |
|---|---|---|
| Loss Aversion | Losses feel ~2x stronger than gains | "Don't lose your saved items" |
| Anchoring | First number influences perception | Show competitor price first |
| Framing Effect | Same info, different presentation | "$3/day" vs "$90/month" |
Social & Identity
| Principle | Definition | Example |
|---|---|---|
| Social Proof | People follow what others do | "12,847 users joined this month" |
| Authority | People trust credible experts | Certifications, credentials, expert endorsements |
Choice Architecture
| Principle | Definition | Example |
|---|---|---|
| Default Effect | People stick with pre-selected options | Pre-select annual plan |
| Choice Overload | Too many options = paralysis | Show 3 plans, not 7 |
| Friction/Sludge | Small barriers dramatically reduce completion | One-click vs multi-step |
Trust & Risk
| Principle | Definition | Example |
|---|---|---|
| Ambiguity Aversion | People avoid unknown probabilities | Clear "What happens next" steps |
| Zero Risk Bias | Preference for eliminating risk entirely | "30-day money-back guarantee" |
Red Flags in Variation Design
| Red Flag | Problem | Fix |
|---|---|---|
| Variants too similar | Weak signal, inconclusive results | Make variants meaningfully distinct |
| No losing scenario | If it "can't lose," you're not testing anything | Each variant should have a plausible failure mode |
Output Format
Use this structure when proposing variations:
| Variant | What User Sees | Principle | Why It Might Win | Why It Might Lose |
|---------|----------------|-----------|------------------|-------------------|
| Control | [Current state] | Baseline | - | - |
| B | [Description] | [Named principle] | [Evidence/logic] | [Risk] |
| C | [Description] | [Named principle] | [Evidence/logic] | [Risk] |
Then ask: "Does this capture meaningful variation? Any variants to add or modify?" and "Do we have enough sample to test this many variatiants in a reasonable timeframe?"
スコア
総合スコア
リポジトリの品質指標に基づく評価
SKILL.mdファイルが含まれている
ライセンスが設定されている
100文字以上の説明がある
GitHub Stars 100以上
3ヶ月以内に更新がある
10回以上フォークされている
オープンIssueが50未満
プログラミング言語が設定されている
1つ以上のタグが設定されている
レビュー
レビュー機能は近日公開予定です