🧪 Xperiment: A/B testing, simplified

• Optimize like a pro. Intuition doesn't count, numbers do.
• Make data-driven decisions, not guesses.

Features

• 🎯 Simple API - Easy to integrate and use
• 💾 Persistent Storage - Uses DeepBase for automatic persistence
• 🎲 Configurable Probabilities - Set custom weight for each variant
• 📊 Built-in Analytics - Track hits/misses and generate effectiveness reports
• 🔄 Singleton Pattern - Ensures consistent user experience
• ⚡ Async/Await - Modern JavaScript API

Installation

npm install xperiment

Demo Examples

Check out the demo folder for complete, runnable examples:

• basic.js - Simple A/B test with two variants
• multivariant.js - Testing 4 variants simultaneously (A/B/C/D)
• weighted-tracking.js - Using weighted scores for different actions
• score-usage.js - Using score() for engagement time tracking
• dashboard.js - Monitoring multiple experiments with a visual dashboard
• complete-flow.js - Multi-stage funnel testing for e-commerce

Run any example:

node demo/basic.js
node demo/score-usage.js
node demo/dashboard.js

Quick Start

Simple Usage (Single Experiment)

import Xperiment from 'xperiment';

// Create experiment directly with cases
const exp = new Xperiment('user123', {
    cases: ['variant_a', 'variant_b']
    // name is optional, defaults to 'default'
});

// Get assigned variant
const variant = await exp.case();
console.log(`User assigned to: ${variant}`);

// Track outcomes
await exp.hit();
await exp.miss();

Production Usage (Multiple Users)

import Xperiment from 'xperiment';

// 1. Define experiment once (persists in database)
await Xperiment.define(['variant_a', 'variant_b'], 'homepage-test');
// Or with custom weights: { variant_a: 30, variant_b: 70 }

// 2. Get experiment instance for each user (loads config from DB)
const exp = await Xperiment.get('user123', 'homepage-test');

// 3. Get the assigned variant (persists automatically)
const variant = await exp.case();

// 4. Track outcomes
await exp.hit(5);    // Add 5 points
await exp.miss(2);   // Subtract 2 points

// 5. Generate effectiveness report
const report = await Xperiment.report('homepage-test');
console.log(`Best variant: ${report.bestCase}`);

// 6. Reset experiment (clears all data)
await Xperiment.reset('homepage-test');

Inline Usage (Quick & Simple)

// Define and get in one step
const exp = await Xperiment.get('user123', 'my-test', ['option_a', 'option_b']);
const variant = await exp.case();

API Reference

Static Method: define()

Define an experiment with its cases. Configuration is persisted in the database.

await Xperiment.define(cases, name = 'default')

Parameters:

• cases (Array|Object) - Case definitions
  • Array: ['option1', 'option2'] - Equal probability (1/n each)
  • Object: {option1: 30, option2: 70} - Custom weights
• name (string) - Experiment name (optional, defaults to 'default')

Example:

// Equal distribution
await Xperiment.define(['headline_a', 'headline_b', 'headline_c'], 'headline-test');

// Custom weights
await Xperiment.define({ red: 30, blue: 70 }, 'button-test');

// Using default name (no need to specify)
await Xperiment.define(['option_a', 'option_b']);

Constructor

Create an experiment instance directly. Ideal for simple use cases.

new Xperiment(id, options)

Parameters:

• id (string) - Unique user identifier
• options (object) - Configuration options
  • name (string) - Experiment name (default: 'default')
  • cases (Array|Object) - Case definitions (optional if loading from DB)

Examples:

// Simple: just cases (uses 'default' name)
const exp1 = new Xperiment('user456', {
    cases: ['red', 'blue']
});

// With custom name and weights
const exp2 = new Xperiment('user456', {
    name: 'button-color-test',
    cases: { red: 30, blue: 70 }
});

// Array with equal probability
const exp3 = new Xperiment('user456', {
    name: 'headline-test',
    cases: ['a', 'b', 'c', 'd']  // 25% each
});

Static Method: get()

Get or create a singleton instance for a user/experiment combination. Automatically loads experiment configuration from database.

await Xperiment.get(id, nameOrOptions = 'default', cases = null)

Parameters:

• id (string) - Unique user identifier
• nameOrOptions (string|Object) - Experiment name or options object
  • As string: 'experiment-name'
  • As object: { name: 'experiment-name', cases: [...] }
• cases (Array|Object) - Optional: cases to define if experiment doesn't exist

Returns: Promise - Experiment instance

Examples:

// Load from DB (experiment must be defined first)
await Xperiment.define(['control', 'treatment'], 'my-test');
const exp1 = await Xperiment.get('user123', 'my-test');

// Inline definition
const exp2 = await Xperiment.get('user123', 'quick-test', ['a', 'b']);

// With options object
const exp3 = await Xperiment.get('user123', {
    name: 'flex-test',
    cases: ['x', 'y', 'z']
});

// Default experiment (no name needed)
const exp4 = await Xperiment.get('user123'); // uses 'default' name

Instance Method: case()

Get the assigned case for this user. Returns the same case on subsequent calls.

await exp.case()

Returns: Promise - The assigned case

Example:

await Xperiment.define(['control', 'treatment'], 'my-test');
const exp = await Xperiment.get('user123', 'my-test');
const variant = await exp.case();
// Returns 'control' or 'treatment' based on configured probabilities
// Always returns the same value for this user

Instance Method: hit()

Record a positive outcome (success).

await exp.hit(amount = 1)

Parameters:

• amount (number) - Points to add (default: 1)

Example:

await exp.hit();    // Add 1 point
await exp.hit(10);  // Add 10 points

Instance Method: miss()

Record a negative outcome (failure).

await exp.miss(amount = 1)

Parameters:

• amount (number) - Points to add (default: 1)

Example:

await exp.miss();   // Add 1 miss
await exp.miss(5);  // Add 5 misses

Instance Method: score()

Set a fixed score value for a user (non-incremental). Unlike hit() which adds to the total, score() sets a specific value that will be added to hits in calculations.

await exp.score(value = 1)

Parameters:

• value (number) - Fixed score value to set (default: 1)

Use cases:

• Engagement time (seconds/minutes)
• Scroll depth percentage (0-100)
• Revenue per user
• Any metric where you track a final accumulated value per user

Example:

// Track time spent on page
const engagementSeconds = 145;
await exp.score(engagementSeconds);

// Track scroll depth
const scrollPercentage = 87;
await exp.score(scrollPercentage);

Note: Each call to score() replaces the previous value (not incremental). The score value is added to hits when generating reports.

Static Method: reset()

Reset an entire experiment, deleting all user data.

await Xperiment.reset(name = 'default')

Parameters:

• name (string) - Experiment name to reset (default: 'default')

Example:

await Xperiment.reset('homepage-test');
await Xperiment.reset(); // Resets 'default' experiment

Static Method: report()

Generate an effectiveness report for an experiment.

await Xperiment.report(name = 'default')

Parameters:

• name (string) - Experiment name (default: 'default')

Returns: Promise - Report with the following structure:

{
    experiment: 'experiment-name',
    totalUsers: 100,
    cases: {
        'variant_a': {
            users: 50,
            totalHits: 300,
            totalMisses: 100,
            netScore: 200,
            successRate: 0.75
        },
        'variant_b': {
            users: 50,
            totalHits: 250,
            totalMisses: 150,
            netScore: 100,
            successRate: 0.625
        }
    },
    bestCase: 'variant_a',
    effectiveness: 100
}

Example:

const report = await Xperiment.report('homepage-test');
console.log(`Total users tested: ${report.totalUsers}`);
console.log(`Winner: ${report.bestCase}`);
console.log(`Success rate: ${report.cases[report.bestCase].successRate * 100}%`);

Usage Examples

Simple Single Experiment

import Xperiment from 'xperiment';

// No need to define or name - just use it!
const exp = new Xperiment('user_alice', {
    cases: ['old_checkout', 'new_checkout']
});

const variant = await exp.case();

// Show appropriate UI
if (variant === 'new_checkout') {
    showNewCheckout();
} else {
    showOldCheckout();
}

// Track conversion
if (userCompletesPurchase()) {
    await exp.hit();
} else {
    await exp.miss();
}

Production with Multiple Users

import Xperiment from 'xperiment';

// Define experiment once (persists in database)
await Xperiment.define(['old_checkout', 'new_checkout'], 'checkout-flow');

async function testUserJourney(userId) {
    // Get experiment instance for user (loads from DB)
    const exp = await Xperiment.get(userId, 'checkout-flow');
    
    const variant = await exp.case();
    
    // Show appropriate UI based on variant
    if (variant === 'new_checkout') {
        showNewCheckout();
    } else {
        showOldCheckout();
    }
    
    // Track conversion
    if (userCompletesPurchase()) {
        await exp.hit();
    } else {
        await exp.miss();
    }
}

Weighted Distribution

// Give 80% of traffic to control, 20% to new feature
await Xperiment.define({ control: 80, new_feature: 20 }, 'feature-rollout');

const exp = await Xperiment.get('user789', 'feature-rollout');
const variant = await exp.case();

Multi-variant Testing (Equal Distribution)

// Define with array for equal probability (25% each)
await Xperiment.define([
    'headline_a', 
    'headline_b', 
    'headline_c', 
    'headline_d'
], 'landing-page-headline');

const exp = await Xperiment.get('user999', 'landing-page-headline');
const headline = await exp.case();

Analytics Dashboard

async function showDashboard() {
    const experiments = ['homepage-test', 'checkout-flow', 'pricing-test'];
    
    for (const name of experiments) {
        const report = await Xperiment.report(name);
        
        console.log(`\n=== ${report.experiment} ===`);
        console.log(`Total Users: ${report.totalUsers}`);
        console.log(`Best Case: ${report.bestCase}`);
        
        for (const [caseName, stats] of Object.entries(report.cases)) {
            console.log(`\n${caseName}:`);
            console.log(`  Users: ${stats.users}`);
            console.log(`  Success Rate: ${(stats.successRate * 100).toFixed(2)}%`);
            console.log(`  Net Score: ${stats.netScore}`);
        }
    }
}

Score-based Tracking

await Xperiment.define(['layout_a', 'layout_b'], 'engagement-test');

const exp = await Xperiment.get('user555', 'engagement-test');
const layout = await exp.case();

// Track different levels of engagement
if (userClicksButton()) {
    await exp.hit(1);
}
if (userSharesContent()) {
    await exp.hit(5);
}
if (userMakesPurchase()) {
    await exp.hit(10);
}
if (userBounces()) {
    await exp.miss(1);
}

Testing

Run the test suite:

npm test

The library includes comprehensive tests covering:

• Constructor and singleton pattern
• Case assignment and persistence
• Metrics tracking
• Reset functionality
• Report generation
• Edge cases and error handling

How It Works

1. Assignment: When a user first encounters an experiment, they're randomly assigned to a case based on configured probabilities
2. Persistence: The assignment is immediately saved to DeepBase and will remain consistent for that user
3. Tracking: As the user interacts with your application, you track positive (hit) and negative (miss) outcomes
4. Analysis: Generate reports to see which variant performs best based on net score (hits - misses) and success rate

Best Practices

1. Choose meaningful experiment names - Use descriptive names like 'homepage-hero-test' instead of 'test1'
2. Track meaningful events - Use hits for conversions, not just clicks
3. Use weighted scoring - Give more points to important actions (e.g., purchase = 10 points, signup = 5 points)
4. Let tests run long enough - Collect sufficient data before making decisions
5. Reset carefully - Resetting an experiment deletes ALL user data for that experiment

Data Structure

DeepBase stores data in the following structure:

config/
  {experimentName}/
    cases: ['variant_a', 'variant_b'] or { variant_a: 50, variant_b: 50 }

experiments/
  {experimentName}/
    {userId}/
      case: 'variant_a'
      hits: 25
      misses: 10
      score: 145  (optional, set via score() method)

License

MIT

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.