What is Criterion?

Criterion is a universal decision engine for business-critical decisions. It provides a declarative, deterministic, and fully explainable way to encode business logic.

The Problem

Business rules are everywhere:

• Should this transaction be flagged as high-risk?
• Is this user eligible for a premium tier?
• What discount applies to this order?
• Should this loan application be approved?

These rules often end up scattered across codebases as nested if/else statements, making them:

• Hard to audit - Where exactly is the logic?
• Hard to test - Too many code paths
• Hard to explain - Why was this decision made?
• Hard to change - Fear of breaking something

The Solution

Criterion provides a structured way to define decisions:

const decision = {
  id: "loan-approval",
  version: "1.0.0",
  inputSchema,     // What data do we need?
  outputSchema,    // What do we return?
  profileSchema,   // What parameters can vary?
  rules: [...]     // The actual logic
};

Each decision is:

• Pure - No side effects, no I/O
• Deterministic - Same input → same output
• Explainable - Every rule has an explain function
• Validated - Zod schemas catch invalid data
• Auditable - Full trace of which rules were evaluated

Core Principles

1. Pure Core

The engine never performs I/O. It receives data, evaluates rules, and returns a result. Fetching data, logging, and persistence happen outside.

2. Mandatory Contracts

Every decision declares its input, output, and profile schemas using Zod. Invalid data is rejected before rule evaluation begins.

3. Determinism

Given the same input and profile, the engine always returns the same result. No randomness, no time-dependent logic, no external state.

4. Explainability

Every decision result includes:

• Which rule matched
• Why it matched (via explain)
• Which rules were evaluated
• The full evaluation trace

5. Small Core

Criterion does one thing: evaluate decisions. It doesn't fetch data, send notifications, or manage workflows. Integrations happen in userland.

When to Use Criterion

Criterion is ideal for:

• Risk assessment - Transaction monitoring, fraud detection
• Eligibility checks - User tiers, feature access, discounts
• Compliance rules - Regulatory requirements, policy enforcement
• Approval workflows - Loan decisions, claim processing
• Dynamic configuration - A/B testing, feature flags with logic

When NOT to Use Criterion

Criterion is not designed for:

• Stateful workflows - Use a workflow engine
• Machine learning - Use ML frameworks
• Complex event processing - Use a CEP system
• Real-time streaming - Use stream processors

Next Steps

• Getting Started - Install and run your first decision
• Core Concepts - Understand the building blocks