Rules engine

What a rule looks like

A rule has:

• A name and optional description.
• A boolean condition tree (AND / OR of leaf comparisons).
• A score impact (signed integer, added to the rules-scorer total when the rule fires).
• A priority (orders the rule list for display only).

Condition shape

{
  "logic": "and",
  "conditions": [
    { "field": "amount", "op": "gt", "value": 500 },
    { "field": "shipping_address.country", "op": "ne", "value": "billing_address.country" }
  ]
}

Supported operators

• eq, ne — equality
• gt, gte, lt, lte — numeric
• in, nin — membership in a list
• contains, starts_with, ends_with — string
• matches — regex (compiled, anchored)
• exists, not_exists — null checks

Fields you can reference

All fields from the /v1/analyze payload, plus derived signals: customer.email_domain, history.prior_orders, history.prior_chargebacks, payment.avs_result, payment.cvv_result, and more. The visual builder in Rules → New rule lists every available field with type hints.

AI rule creator

Describe a fraud pattern in plain English on any paid tier:

“Flag orders over $500 from new emails that ship to a different country than the billing address.”

The AI generates a structured rule and shows you the resulting JSON before saving. It also calibrates the score impact against your historical data, so the rule doesn't over- or under-fire on day one.

Profile presets

Four built-in profiles let you start from a tested ruleset and tune:

• Base — five high-confidence rules. Right for low-risk verticals (apparel, accessories).
• Medium — adds velocity and address-mismatch checks.
• Advanced — adds BIN, IP, device, and history rules.
• Expert — the full library plus aggressive thresholds. For high-AOV / digital-good stores.

Shadow vs live

Rules fire identically in shadow and live modes. Only side effects differ (no Order Risk writeback, no auto-cancel in shadow). This is by design — shadow gives you a faithful preview of what your live ruleset would do.

Auditing

Every rule firing is recorded on the transaction with the rule's name, condition hash, and score impact at the time of evaluation. Editing a rule does not retroactively change historical assessments.