Decision Profiles

What is a Profile?

A Decision Profile defines how a decision interprets a given context.

• Context describes facts.
• Profile describes thresholds/sensitivity.
• Decisions remain pure and reusable.

Profiles are injected explicitly at runtime:

engine.run(decision, context, { profile: "high-inflation" })

Why not merge profile into context?

Because a profile is not a fact. It is an interpretation layer, and must remain explicit.

What belongs in a profile?

• thresholds (numbers)
• sensitivity modes (conservative/aggressive)
• domain-specific calibrations

What does not belong in a profile?

• raw data (facts)
• side effects
• anything time-dependent unless explicitly injected

Profile Resolution

Criterion does not load profiles by itself. The host system is responsible for profile registration and resolution.

Canonical Flow (ProfileRegistry)

const registry = new ProfileRegistry()

registry.register("high-inflation", {
  high_risk_inflation_threshold: 0.05,
  high_risk_horizon_days: 60,
  medium_risk_inflation_threshold: 0.03,
  medium_risk_horizon_days: 30
})

engine.run(decision, context, { profile: "high-inflation" }, registry)

• The engine never loads JSON or reads from disk
• Registry lives outside the core
• Profile ID is resolved to concrete values before evaluation

Escape Hatch: Inline Profile Object

For testing, simulations, or research:

engine.run(decision, context, {
  profile: {
    high_risk_inflation_threshold: 0.05,
    high_risk_horizon_days: 60,
    medium_risk_inflation_threshold: 0.03,
    medium_risk_horizon_days: 30
  }
})

This is allowed but not recommended as the default pattern.

Profile Validation

The engine validates that the provided profile contains all parameters declared in the decision's <profile-parameters> schema.

Missing or mistyped parameters result in INVALID_INPUT (profile invalid).

Success criteria

• The same decision can be reused globally by swapping profiles
• The engine remains domain-agnostic
• Explainability includes the profile used