@splitlab/core

Shared evaluation engine for SplitLab SDKs. Pure functions, zero dependencies.

What's in here

• localEvaluate() — evaluate all experiments and flags for a user in < 1ms
• evaluateRules() — targeting rule engine (supports is, contains, gt, in, segments, etc.)
• Deterministic hashing — consistent experiment bucketing and flag rollout
• parseUserAgent() — lightweight UA parser (browser, OS, device type)
• Types — shared TypeScript types used across all SDKs (ExperimentConfig, FlagConfig, ServerConfig, EvalResult, etc.)

Installation

npm install @splitlab/core

Usage

import { localEvaluate } from '@splitlab/core';
import type { ServerConfig, EvalResult } from '@splitlab/core';

// Evaluate all experiments and flags locally
const config: ServerConfig = await fetchConfigFromServer();
const result: EvalResult = localEvaluate(config, 'user-123', { country: 'US' });

console.log(result.experiments); // { 'checkout-btn': 'variant_a' }
console.log(result.flags);       // { 'dark-mode': true }

Targeting Rules

The rule engine supports nested conditions with all/any matching:

import { evaluateRules } from '@splitlab/core';

const rules = {
  match: 'all',
  groups: [{
    conditions: [
      { attribute: 'country', operator: 'is', value: 'US' },
      { attribute: 'plan', operator: 'in', value: ['pro', 'enterprise'] },
    ],
  }],
};

evaluateRules(rules, { country: 'US', plan: 'pro' }, []);
// → true

Operators: is, is_not, contains, not_contains, gt, lt, gte, lte, in, not_in.

Building

npm run build   # ESM + CJS + types → dist/
npm run dev     # Watch mode

All evaluation logic lives here so that browser, server, and React SDKs produce identical results.