Compliance engine · LendEasy Docs

Resolve current facts and effective policy into an explainable, expiring decision before a contact, disclosure, AI tool call, or account action can execute.

The compliance engine is a decision boundary, not a warning banner. It evaluates one concrete proposed action for one subject at one time. A blocked action does not execute; a missing required fact is not treated as permission.

Decision contract

Outcome	Meaning	Executor behavior
`allowed`	All applicable rules passed on sufficiently fresh facts.	May proceed before expiry if authorization and state still hold.
`allowed_with_warning`	No prohibition matched, but configured risk or proximity threshold exists.	Proceed only under the warning’s declared policy.
`approval_required`	A permitted human must accept the frozen action and evidence.	Hold; creator cannot self-approve where separation applies.
`blocked`	At least one non-overridable rule prohibits this path.	Do not execute; record blocked outcome and next safe route.
`missing_facts`	A required fact is absent, contradictory, or stale.	Refresh, collect evidence, or route to review.

Outcomes include expiresAt and a decision ID. Expiry is deliberately short for contact, restriction, and money-sensitive actions.

Evaluate a proposed action

curl -X POST "$BASE/v1/servicing/compliance/evaluations" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: evaluate-harbor-call-2027-05-01T101158" \
  -d '{
    "action": "send_interaction",
    "subject": {
      "customerId": "412",
      "loanIds": ["7204"]
    },
    "parameters": {
      "channel": "voice",
      "purpose": "collections",
      "contactPointId": "cp_MAYA_MOBILE_1"
    },
    "asOf": "2027-05-01T10:11:58-07:00"
  }'

{
  "decisionId": "dec_CONTACT_81",
  "outcome": "allowed",
  "expiresAt": "2027-05-01T10:16:58-07:00",
  "rules": [
    { "ruleId": "contact-time-CA-v4", "outcome": "pass" }
  ],
  "factsFreshness": "fresh"
}

Resolution pipeline

typed action
   ↓
scope resolution ─ tenant · product · debt · purpose · channel · jurisdiction
   ↓
fact resolution  ─ value · source · asOf · observedAt · freshness · confidence
   ↓
effective rules ─ authority · applicability · exceptions · version · dates
   ↓
rule results     ─ pass · warn · approval · block · missing
   ↓
aggregate + persist + expire

The aggregator does not use a simplistic numerical score. A scoped hard stop dominates a general allow; missing input to a required rule prevents an allow; an approval requirement cannot be downgraded by a lower-risk rule.

Fact contract

Every fact used in a durable decision records:

canonical name and typed value;
subject and scope;
source and source reference;
source asOf and LendEasy observedAt times;
freshness classification and maximum age;
confidence or verification state where relevant.

Balances can tolerate a different freshness budget than bankruptcy, cease, consent, or payment-in-flight state. Bring your own core integrations publish these budgets explicitly.

Planning versus execution

A direct evaluation helps a UI present available choices. It does not let the caller bypass the action-owning API. POST /v1/servicing/interactions, action submission, and other execution endpoints perform their own last-moment decision and link it to the result.

Boundaries

The engine can apply versioned controls approved for a tenant; it does not decide which law applies, replace legal review, verify that policy text is legally sufficient, or turn a product label into a regulatory conclusion. Those are governance responsibilities represented in the rule lifecycle.

Never implement “engine unavailable means allow.” A decision-dependent action fails closed or follows its explicit degraded-mode route when facts, policies, or the decision service are unavailable.