Case Intelligence API

Decision support endpoint aggregating AI assessment, similar cases, applicable rules, and officer corrections for a specific compliance case.

Endpoints

Get Case Intelligence

GET /api/cases/{workflow_id}/case-intelligence
GET /api/cases/{workflow_id}/case-intelligence?retry=true

Query Parameters:

Parameter	Type	Default	Description
retry	boolean	false	Clear cached assessment and re-trigger generation

Response:

{
  "assessment": {
    "summary": "Based on 3 similar Belgian PSP cases, this case aligns with the approved pattern. Key risk: single-director control structure matches 2 prior escalations.",
    "pattern": "approved",
    "key_differences": [
      "Revenue decline exceeds threshold seen in approved cases",
      "Single-director pattern flagged in 2 of 3 precedents"
    ],
    "assessment_confidence": 0.82,
    "similar_count": 3,
    "approved_count": 2,
    "generated_at": "2026-03-17T14:23:00Z"
  },
  "assessment_status": "ready",
  "similar_cases": [
    {
      "workflow_id": "wf-abc-123",
      "company_name": "FinTech Solutions BV",
      "registration_number": "BE0123456789",
      "country": "BE",
      "template_type": "PSP Merchant",
      "risk_score": 0.35,
      "outcome": "approved",
      "decided_at": "2026-03-10T09:15:00Z",
      "similarity_score": 0.87,
      "key_shared_factors": ["BE", "PSP Merchant"]
    }
  ],
  "applicable_rules": [
    {
      "text": "For Belgian management companies where a single natural person occupies all key roles, always require a live interview before approval.",
      "scope": "BE management companies",
      "trigger": "sole natural person in all key roles",
      "action": "create live interview task",
      "risk_rationale": "Single-person control structures elevate risk",
      "source": "officer-taught",
      "created_at": "2026-02-15T10:30:00Z"
    }
  ],
  "officer_corrections": [
    {
      "id": "corr-456",
      "finding_category": "financial_health",
      "signal_type": "finding_rejected",
      "rejection_reason": "Revenue decline is seasonal, not structural",
      "description": "Revenue declined 34% YoY — HIGH risk signal",
      "created_at": "2026-03-16T11:00:00Z",
      "has_analysis": true
    }
  ]
}

Assessment Status Values:

Status	Meaning	Frontend Behavior
ready	Assessment cached and available	Display assessment
generating	Background task running	Show spinner, poll every 2s
failed	Generation failed	Show retry button

Authentication: Bearer JWT token required. Officer must have access to the case's tenant.

Multi-tenancy: All database queries use get_tenant_session(tenant_id) for RLS enforcement.

Response Types

CaseAssessment

Field	Type	Description
summary	string	2-4 sentence prose assessment
pattern	enum	approved, rejected, mixed, insufficient
key_differences	string[]	Max 3 differences that could change outcome
assessment_confidence	float	0-1, distinct from 4-dimension confidence scoring
similar_count	int	Number of similar cases considered
approved_count	int	Number approved among similar cases
generated_at	string	ISO timestamp

SimilarCaseMatch

Field	Type	Description
workflow_id	string	Deep-linkable case identifier
company_name	string	Real company name from cases table
registration_number	string	Enterprise/registration number
country	string	ISO country code
template_type	string	Workflow template ID
risk_score	float?	From confidence_score JSONB
outcome	enum	approved, rejected, follow_up, escalated, pending
decided_at	string?	ISO timestamp of latest decision
similarity_score	float	0-1 from Letta archival search
key_shared_factors	string[]	Common attributes

CaseIntelligenceRule

Field	Type	Description
text	string	Rule description
scope	string	Applicability scope
trigger	string?	Activation condition
action	string?	Required response
risk_rationale	string?	Why this rule exists
source	enum	auto-learned or officer-taught
created_at	string?	ISO timestamp

OfficerCorrection

Field	Type	Description
id	string	Correction identifier
finding_category	string	Risk domain category
signal_type	enum	finding_confirmed or finding_rejected
rejection_reason	string?	Why finding was rejected
description	string	Original finding text
created_at	string	ISO timestamp
has_analysis	boolean	Whether FindingAnalysis exists