{
  "skill_name": "ab-testing",
  "evals": [
    {
      "id": 1,
      "prompt": "I want to A/B test our homepage headline. We currently say 'The All-in-One Project Management Tool' and want to test something benefit-focused. We get about 15,000 visitors/month and our current signup rate is 3.2%.",
      "expected_output": "Should check for product-marketing.md first. Should build a proper hypothesis using the framework: 'Because [observation], we believe [change] will cause [outcome], which we'll measure by [metric].' Should identify this as an A/B test (two variants). Should calculate or reference sample size needs based on 15,000 monthly visitors and 3.2% baseline. Should define primary metric (signup rate), secondary metrics, and guardrail metrics. Should warn about the peeking problem and recommend a fixed test duration. Should provide the test plan in the structured output format.",
      "assertions": [
        "Checks for product-marketing.md",
        "Uses the hypothesis framework with observation, belief, outcome, and metric",
        "Identifies as A/B test type",
        "Addresses sample size calculation based on traffic and baseline rate",
        "Defines primary metric (signup rate)",
        "Defines secondary and guardrail metrics",
        "Warns about the peeking problem",
        "Provides structured test plan output"
      ],
      "files": []
    },
    {
      "id": 2,
      "prompt": "we want to test like 4 different CTA button colors on our pricing page. is that a good idea?",
      "expected_output": "Should trigger on casual phrasing. Should identify this as an A/B/n test (multiple variants). Should caution that testing 4 variants requires significantly more traffic than a simple A/B test. Should reference the sample size quick reference showing traffic multipliers for multiple variants. Should question whether button color alone is likely to produce meaningful lift vs testing CTA copy, placement, or surrounding context. Should recommend either reducing to 2 variants or ensuring sufficient traffic. Should still provide hypothesis framework and test setup if proceeding.",
      "assertions": [
        "Triggers on casual phrasing",
        "Identifies as A/B/n test (multiple variants)",
        "Cautions about increased traffic needs for 4 variants",
        "References sample size requirements",
        "Questions whether button color alone is high-impact",
        "Suggests alternative higher-impact elements to test",
        "Provides hypothesis framework"
      ],
      "files": []
    },
    {
      "id": 3,
      "prompt": "Our test has been running for 3 days and Variant B is winning with 95% confidence. Should we call it?",
      "expected_output": "Should immediately address the peeking problem. Should explain that checking results early inflates false positive rates. Should recommend running for the full pre-calculated duration regardless of early results. Should explain why early significance can be misleading (regression to the mean, day-of-week effects, audience mix shifts). Should provide guidance on when it IS appropriate to stop early (sequential testing methods). Should recommend the pre-test commitment to duration.",
      "assertions": [
        "Addresses the peeking problem directly",
        "Explains why early significance is misleading",
        "Recommends running for full pre-calculated duration",
        "Mentions day-of-week effects or audience mix shifts",
        "Explains false positive rate inflation from peeking",
        "Mentions sequential testing as alternative approach"
      ],
      "files": []
    },
    {
      "id": 4,
      "prompt": "Help me set up a multivariate test on our landing page. I want to test the headline, hero image, and CTA button simultaneously.",
      "expected_output": "Should identify this as a Multivariate Test (MVT). Should explain that MVT tests combinations of elements and requires much more traffic than A/B tests. Should calculate or reference traffic needs (combinations multiply: e.g., 2 headlines × 2 images × 2 CTAs = 8 combinations). Should recommend MVT only if traffic supports it, otherwise suggest sequential A/B tests. Should build hypotheses for each element being tested. Should define interaction effects to watch for. Should provide structured test plan.",
      "assertions": [
        "Identifies as multivariate test (MVT)",
        "Explains MVT tests combinations of elements",
        "Addresses dramatically higher traffic requirements",
        "Calculates number of combinations",
        "Suggests sequential A/B tests as alternative if traffic insufficient",
        "Builds hypotheses for each element",
        "Provides structured test plan"
      ],
      "files": []
    },
    {
      "id": 5,
      "prompt": "What metrics should I track for an A/B test on our trial signup page? We're testing a longer form (adds company size and role fields) against the current short form.",
      "expected_output": "Should apply the metrics selection framework with three tiers: primary, secondary, and guardrail metrics. Primary: form completion rate (the direct conversion metric). Secondary: lead quality metrics (SQL conversion rate, activation rate post-signup). Guardrail: overall signup volume (ensure longer form doesn't tank total signups below acceptable threshold). Should explain the tradeoff between conversion quantity and lead quality. Should note that this test needs longer observation window to measure downstream metrics.",
      "assertions": [
        "Applies three-tier metric framework (primary, secondary, guardrail)",
        "Identifies form completion rate as primary metric",
        "Identifies lead quality as secondary metric",
        "Defines guardrail metrics to protect against negative outcomes",
        "Explains quantity vs quality tradeoff",
        "Notes need for longer observation window for downstream metrics"
      ],
      "files": []
    },
    {
      "id": 6,
      "prompt": "Can you help me write copy for our new landing page? We want to test it against the current version.",
      "expected_output": "Should recognize this is primarily a copywriting task, not a test setup task. Should defer to or cross-reference the copywriting skill for writing the actual copy. May help frame the test hypothesis and setup, but should make clear that copywriting is the right skill for creating the page copy itself.",
      "assertions": [
        "Recognizes this as primarily a copywriting task",
        "References or defers to copywriting skill",
        "Does not attempt to write full page copy using test setup patterns",
        "May offer to help with test hypothesis and setup"
      ],
      "files": []
    },
    {
      "id": 7,
      "prompt": "We ran an A/B test on our pricing page for 4 weeks. Control: 2.1% conversion. Variant: 2.4% conversion. 12,000 visitors per variant. Is this statistically significant? Should we ship it?",
      "expected_output": "Should calculate the result under a declared analysis: 252/12,000 vs 288/12,000 gives approximately p = 0.117 for an uncorrected two-sided pooled-proportion z-test, so it does not meet alpha = 0.05. Should distinguish the 0.3 percentage-point / ~14% relative estimate from evidence for a reliably positive effect. Report uncertainty and practical business impact. Respect the pre-specified duration and stopping rule; do not recommend continuing the completed fixed-horizon test until significant. Unplanned segments are exploratory; a follow-up test needs its own plan.",
      "assertions": [
        "Uses actual conversion counts and states the analysis assumptions",
        "Does not call p ~0.117 significant at alpha = 0.05",
        "Distinguishes statistical uncertainty from business value",
        "Does not extend a completed fixed-horizon test until significant",
        "Labels unplanned segment findings exploratory"
      ],
      "files": []
    },
    {
      "id": 8,
      "prompt": "Our fixed-horizon A/B test reached its planned sample today with p = 0.04. Can I tell the CEO there is a 96% chance the change works and just a 4% chance the lift was random? We have a +2% estimate, an interval of +0.1% to +3.9%, and a higher refund rate in the variant.",
      "expected_output": "Reject both probability interpretations. Explain the p-value conditional on the null model and assumptions, and repeated-sampling interval coverage without asserting a posterior probability. Consider the effect interval, practical impact and refund guardrail before recommending rollout. Do not substitute a numerical significance threshold for the business decision.",
      "assertions": [
        "Does not interpret 1-p as the probability the variant works",
        "Does not interpret p as the probability the result was random",
        "Explains the conditional null-model meaning",
        "Uses interval, business effect and refund guardrail before rollout"
      ],
      "files": []
    }
  ]
}

# Sample Size Guide

Reference for calculating sample sizes and test duration.

## Contents
- Sample Size Fundamentals (required inputs, what these mean)
- Sample Size Quick Reference Tables
- Duration Calculator (formula, examples, minimum duration rules, maximum duration guidelines)
- Online Calculators
- Adjusting for Multiple Variants
- Common Sample Size Mistakes
- When Sample Size Requirements Are Too High
- Sequential Testing
- Quick Decision Framework

## Sample Size Fundamentals

### Required Inputs

1. **Baseline conversion rate**: Your current rate
2. **Minimum detectable effect (MDE)**: Smallest change worth detecting
3. **Statistical significance level**: Usually α = 0.05 (5%)
4. **Statistical power**: Usually 80% (β = 0.20)

### What These Mean

**Baseline conversion rate**: If your page converts at 5%, that's your baseline.

**MDE (Minimum Detectable Effect)**: The smallest improvement you care about detecting. Set this based on:
- Business impact (is a 5% lift meaningful?)
- Implementation cost (worth the effort?)
- Realistic expectations (what have past tests shown?)

**Significance level (α = 0.05)**: For the planned test and its assumptions, limits the long-run probability of rejecting a true null hypothesis to 5%. It is not the probability that an observed difference is “due to chance.” Repeated unadjusted looks or multiple comparisons do not retain that error rate. See the [ASA's interpretation guidance](https://www.amstat.org/asa/files/pdfs/p-valuestatement.pdf).

**Statistical power (80%)**: Means if there's a real effect of size MDE, you have 80% chance of detecting it.

---

## Sample Size Quick Reference Tables

### Conversion Rate: 1%

| Lift to Detect | Sample per Variant | Total Sample |
|----------------|-------------------|--------------|
| 5% (1% → 1.05%) | 1,500,000 | 3,000,000 |
| 10% (1% → 1.1%) | 380,000 | 760,000 |
| 20% (1% → 1.2%) | 97,000 | 194,000 |
| 50% (1% → 1.5%) | 16,000 | 32,000 |
| 100% (1% → 2%) | 4,200 | 8,400 |

### Conversion Rate: 3%

| Lift to Detect | Sample per Variant | Total Sample |
|----------------|-------------------|--------------|
| 5% (3% → 3.15%) | 480,000 | 960,000 |
| 10% (3% → 3.3%) | 120,000 | 240,000 |
| 20% (3% → 3.6%) | 31,000 | 62,000 |
| 50% (3% → 4.5%) | 5,200 | 10,400 |
| 100% (3% → 6%) | 1,400 | 2,800 |

### Conversion Rate: 5%

| Lift to Detect | Sample per Variant | Total Sample |
|----------------|-------------------|--------------|
| 5% (5% → 5.25%) | 280,000 | 560,000 |
| 10% (5% → 5.5%) | 72,000 | 144,000 |
| 20% (5% → 6%) | 18,000 | 36,000 |
| 50% (5% → 7.5%) | 3,100 | 6,200 |
| 100% (5% → 10%) | 810 | 1,620 |

### Conversion Rate: 10%

| Lift to Detect | Sample per Variant | Total Sample |
|----------------|-------------------|--------------|
| 5% (10% → 10.5%) | 130,000 | 260,000 |
| 10% (10% → 11%) | 34,000 | 68,000 |
| 20% (10% → 12%) | 8,700 | 17,400 |
| 50% (10% → 15%) | 1,500 | 3,000 |
| 100% (10% → 20%) | 400 | 800 |

### Conversion Rate: 20%

| Lift to Detect | Sample per Variant | Total Sample |
|----------------|-------------------|--------------|
| 5% (20% → 21%) | 60,000 | 120,000 |
| 10% (20% → 22%) | 16,000 | 32,000 |
| 20% (20% → 24%) | 4,000 | 8,000 |
| 50% (20% → 30%) | 700 | 1,400 |
| 100% (20% → 40%) | 200 | 400 |

---

## Duration Calculator

### Formula

```
Duration (days) = (Sample per variant × Number of variants) / (Daily traffic × % exposed)

For unequal allocation, calculate each arm separately:
Days per arm = Required sample for that arm / (Daily eligible traffic × Exposure fraction × Arm allocation fraction)
Total collection time = longest arm duration
```

Use fractions for exposure and allocation (50% = 0.5). Count eligible randomized units consistently with the sample-size calculation; repeat pageviews are not additional participants. The balanced formula assumes equal sample requirements and equal allocation.

### Examples

**Scenario 1: High-traffic page**
- Need: 10,000 per variant (2 variants = 20,000 total)
- Daily traffic: 5,000 visitors
- 100% exposed to test
- Duration: 20,000 / 5,000 = **4 days**

**Scenario 2: Medium-traffic page**
- Need: 30,000 per variant (60,000 total)
- Daily traffic: 2,000 visitors
- 100% exposed
- Duration: 60,000 / 2,000 = **30 days**

**Scenario 3: Low-traffic with partial exposure**
- Need: 15,000 per variant (30,000 total)
- Daily traffic: 500 visitors
- 50% exposed to test
- Effective daily: 250
- Duration: 30,000 / 250 = **120 days** (too long!)

### Minimum Duration Rules

Even with sufficient sample size, run tests for at least:
- **1 full week**: To capture day-of-week variation
- **2 business cycles**: If B2B (weekday vs. weekend patterns)
- **Through paydays**: If e-commerce (beginning/end of month)

### Maximum Duration Guidelines

Avoid running tests longer than 4-8 weeks:
- Novelty effects wear off
- External factors intervene
- Opportunity cost of other tests

---

## Online Calculators

### Recommended Tools

**Evan Miller's Calculator**
https://www.evanmiller.org/ab-testing/sample-size.html
- Simple interface
- Bookmark-worthy

**Optimizely's Calculator**
https://www.optimizely.com/sample-size-calculator/
- Business-friendly language
- Duration estimates

**AB Test Guide Calculator**
https://www.abtestguide.com/calc/
- Includes Bayesian option
- Multiple test types

**VWO Duration Calculator**
https://vwo.com/tools/ab-test-duration-calculator/
- Duration-focused
- Good for planning

---

## Adjusting for Multiple Variants

With more than 2 variants (A/B/n tests), you need more sample:

| Variants | Multiplier |
|----------|------------|
| 2 (A/B) | 1x |
| 3 (A/B/C) | ~1.5x |
| 4 (A/B/C/D) | ~2x |
| 5+ | Consider reducing variants |

**Why?** More comparisons increase chance of false positives. You're comparing:
- A vs B
- A vs C
- B vs C (sometimes)

Apply Bonferroni correction or use tools that handle this automatically.

---

## Common Sample Size Mistakes

### 1. Underpowered tests
**Problem**: Not enough sample to detect realistic effects
**Fix**: Be realistic about MDE, get more traffic, or don't test

### 2. Overpowered tests
**Problem**: Waiting for sample size when you already have significance
**Fix**: This is actually fine—you committed to sample size, honor it

### 3. Wrong baseline rate
**Problem**: Using wrong conversion rate for calculation
**Fix**: Use the specific metric and page, not site-wide averages

### 4. Ignoring segments
**Problem**: Calculating for full traffic, then analyzing segments
**Fix**: If you plan segment analysis, calculate sample for smallest segment

### 5. Testing too many things
**Problem**: Dividing traffic too many ways
**Fix**: Prioritize ruthlessly, run fewer concurrent tests

---

## When Sample Size Requirements Are Too High

Options when you can't get enough traffic:

1. **Increase MDE**: Accept only detecting larger effects (20%+ lift)
2. **Lower confidence**: Use 90% instead of 95% (risky, document it)
3. **Reduce variants**: Test only the most promising variant
4. **Combine traffic**: Test across multiple similar pages
5. **Test upstream**: Test earlier in funnel where traffic is higher
6. **Don't test**: Make decision based on qualitative data instead
7. **Longer test**: Accept longer duration (weeks/months)

---

## Sequential Testing

If you must check results before reaching sample size:

### What is it?
Statistical method that adjusts for multiple looks at data.

### When to use
- High-risk changes
- Need to stop bad variants early
- Time-sensitive decisions

### Tools that support it
- Optimizely (Stats Accelerator)
- VWO (SmartStats)
- PostHog (Bayesian approach)

### Tradeoff
- More flexibility to stop early
- Slightly larger sample size requirement
- More complex analysis

---

## Quick Decision Framework

### Can I run this test?

```
Daily eligible traffic (randomization units): _____
Fraction exposed to the experiment: _____
Number of arms (including control): _____
Allocation fraction for each arm: _____
Baseline conversion rate: _____
MDE I care about: _____

Sample needed per variant: _____ (from tables above)
Balanced allocation: Days = Sample per arm × Number of arms / (Daily eligible traffic × Exposure fraction)
Unequal allocation: Days = max across arms of [Sample required for arm / (Daily eligible traffic × Exposure fraction × Arm allocation)]
Round collection time up, then apply the minimum-duration rules above.

If days > 60: Consider alternatives
If days > 30: Acceptable for high-impact tests
If days < 14: Likely feasible
If days < 7: Easy to run, consider running longer anyway
```

**Duration check (illustrative):** with 2,000 participants required per arm and 1,000 eligible participants/day, two arms at 50/50 with full exposure require four days, not two. At 50% exposure they require eight days. With full exposure and 90/10 allocation, the smaller arm needs 20 days to reach 2,000. These are collection estimates; honor planned business cycles and the analysis stopping rule.

# A/B Test Templates Reference

Templates for planning, documenting, and analyzing experiments.

## Contents
- Test Plan Template
- Results Documentation Template
- Test Repository Entry Template
- Quick Test Brief Template
- Stakeholder Update Template
- Experiment Prioritization Scorecard
- Hypothesis Bank Template

## Test Plan Template

```markdown
# A/B Test: [Name]

## Overview
- **Owner**: [Name]
- **Test ID**: [ID in testing tool]
- **Page/Feature**: [What's being tested]
- **Planned dates**: [Start] - [End]

## Hypothesis

Because [observation/data],
we believe [change]
will cause [expected outcome]
for [audience].
We'll know this is true when [metrics].

## Test Design

| Element | Details |
|---------|---------|
| Test type | A/B / A/B/n / MVT |
| Duration | X weeks |
| Sample size | X per variant |
| Traffic allocation | 50/50 |
| Tool | [Tool name] |
| Implementation | Client-side / Server-side |

## Variants

### Control (A)
[Screenshot]
- Current experience
- [Key details about current state]

### Variant (B)
[Screenshot or mockup]
- [Specific change #1]
- [Specific change #2]
- Rationale: [Why we think this will win]

## Metrics

### Primary
- **Metric**: [metric name]
- **Definition**: [how it's calculated]
- **Current baseline**: [X%]
- **Minimum detectable effect**: [X%]

### Secondary
- [Metric 1]: [what it tells us]
- [Metric 2]: [what it tells us]
- [Metric 3]: [what it tells us]

### Guardrails
- [Metric that shouldn't get worse]
- [Another safety metric]

## Segment Analysis Plan
- Mobile vs. desktop
- New vs. returning visitors
- Traffic source
- [Other relevant segments]

## Success Criteria
- Winner: [Primary metric improves by X% with 95% confidence]
- Loser: [Primary metric decreases significantly]
- Inconclusive: [What we'll do if no significant result]

## Pre-Launch Checklist
- [ ] Hypothesis documented and reviewed
- [ ] Primary metric defined and trackable
- [ ] Sample size calculated
- [ ] Test duration estimated
- [ ] Variants implemented correctly
- [ ] Tracking verified in all variants
- [ ] QA completed on all variants
- [ ] Stakeholders informed
- [ ] Calendar hold for analysis date
```

---

## Results Documentation Template

```markdown
# A/B Test Results: [Name]

## Summary
| Element | Value |
|---------|-------|
| Test ID | [ID] |
| Dates | [Start] - [End] |
| Duration | X days |
| Result | Winner / Loser / Inconclusive |
| Decision | [What we're doing] |

## Hypothesis (Reminder)
[Copy from test plan]

## Results

### Sample Size
| Variant | Target | Actual | % of target |
|---------|--------|--------|-------------|
| Control | X | Y | Z% |
| Variant | X | Y | Z% |

### Primary Metric: [Metric Name]
| Variant | Value | 95% CI | vs. Control |
|---------|-------|--------|-------------|
| Control | X% | [X%, Y%] | — |
| Variant | X% | [X%, Y%] | +X% |

**Statistical significance**: p = X.XX (95% = sig / not sig)
**Practical significance**: [Is this lift meaningful for the business?]

### Secondary Metrics

| Metric | Control | Variant | Change | Significant? |
|--------|---------|---------|--------|--------------|
| [Metric 1] | X | Y | +Z% | Yes/No |
| [Metric 2] | X | Y | +Z% | Yes/No |

### Guardrail Metrics

| Metric | Control | Variant | Change | Concern? |
|--------|---------|---------|--------|----------|
| [Metric 1] | X | Y | +Z% | Yes/No |

### Segment Analysis

**Mobile vs. Desktop**
| Segment | Control | Variant | Lift |
|---------|---------|---------|------|
| Mobile | X% | Y% | +Z% |
| Desktop | X% | Y% | +Z% |

**New vs. Returning**
| Segment | Control | Variant | Lift |
|---------|---------|---------|------|
| New | X% | Y% | +Z% |
| Returning | X% | Y% | +Z% |

## Interpretation

### What happened?
[Explanation of results in plain language]

### Why do we think this happened?
[Analysis and reasoning]

### Caveats
[Any limitations, external factors, or concerns]

## Decision

**Winner**: [Control / Variant]

**Action**: [Implement variant / Keep control / Re-test]

**Timeline**: [When changes will be implemented]

## Learnings

### What we learned
- [Key insight 1]
- [Key insight 2]

### What to test next
- [Follow-up test idea 1]
- [Follow-up test idea 2]

### Impact
- **Projected lift**: [X% improvement in Y metric]
- **Business impact**: [Revenue, conversions, etc.]
```

---

## Test Repository Entry Template

For tracking all tests in a central location:

```markdown
| Test ID | Name | Page | Dates | Primary Metric | Result | Lift | Link |
|---------|------|------|-------|----------------|--------|------|------|
| 001 | Hero headline test | Homepage | 1/1-1/15 | CTR | Winner | +12% | [Link] |
| 002 | Pricing table layout | Pricing | 1/10-1/31 | Plan selection | Loser | -5% | [Link] |
| 003 | Signup form fields | Signup | 2/1-2/14 | Completion | Inconclusive | +2% | [Link] |
```

---

## Quick Test Brief Template

For simple tests that don't need full documentation:

```markdown
## [Test Name]

**What**: [One sentence description]
**Why**: [One sentence hypothesis]
**Metric**: [Primary metric]
**Duration**: [X weeks]
**Result**: [TBD / Winner / Loser / Inconclusive]
**Learnings**: [Key takeaway]
```

---

## Stakeholder Update Template

```markdown
## A/B Test Update: [Name]

**Status**: Running / Complete
**Days remaining**: X (or complete)
**Current sample**: X% of target

### Preliminary observations
[What we're seeing - without making decisions yet]

### Next steps
[What happens next]

### Timeline
- [Date]: Analysis complete
- [Date]: Decision and recommendation
- [Date]: Implementation (if winner)
```

---

## Experiment Prioritization Scorecard

For deciding which tests to run:

| Factor | Weight | Test A | Test B | Test C |
|--------|--------|--------|--------|--------|
| Potential impact | 30% | | | |
| Confidence in hypothesis | 25% | | | |
| Ease of implementation | 20% | | | |
| Risk if wrong | 15% | | | |
| Strategic alignment | 10% | | | |
| **Total** | | | | |

Scoring: 1-5 (5 = best)

---

## Hypothesis Bank Template

For collecting test ideas:

```markdown
| ID | Page/Area | Observation | Hypothesis | Potential Impact | Status |
|----|-----------|-------------|------------|------------------|--------|
| H1 | Homepage | Low scroll depth | Shorter hero will increase scroll | High | Testing |
| H2 | Pricing | Users compare plans | Comparison table will help | Medium | Backlog |
| H3 | Signup | Drop-off at email | Social login will increase completion | Medium | Backlog |
```