CARF Query Processor Skill

Name: Query Processor
Rating: 62
Author: eljaplacido

Purpose

Process analytical queries through the full CARF 4-layer cognitive pipeline:

•Router → Classifies query into Cynefin domain
•Cognitive Mesh → Routes to appropriate domain agent
•Reasoning Services → Causal/Bayesian analysis
•Guardian → Policy enforcement and human escalation

When to Use

•When users ask causal questions ("Why did X cause Y?")
•Uncertainty quantification requests ("How confident are we?")
•Decision-support queries ("Should we take action X?")
•Any query requiring the full CARF reasoning chain

Prerequisites

•API server running: uvicorn src.main:app --port 8000
•.env configured with DEEPSEEK_API_KEY or OPENAI_API_KEY

Execution Steps

1. Submit Query via API

bash

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Why did our costs increase by 15%?",
    "context": {
      "domain": "finance",
      "time_period": "Q4-2025"
    }
  }'

2. With Scenario Payload

Load a pre-configured scenario:

bash

# Get available scenarios
curl http://localhost:8000/scenarios

# Get scenario payload
curl http://localhost:8000/scenarios/scope3_attribution

# Submit with scenario context
curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What causes supplier emissions to increase?",
    "scenario_id": "scope3_attribution"
  }'

3. With Dataset Selection

For causal estimation with real data:

bash

curl -X POST http://localhost:8000/query \
  -H "Content-Type: application/json" \
  -d '{
    "query": "What is the effect of discount on churn?",
    "dataset_selection": {
      "dataset_id": "demo_churn",
      "treatment": "discount_applied",
      "outcome": "churned",
      "covariates": ["tenure", "monthly_charges"]
    }
  }'

Response Schema

json

{
  "session_id": "uuid",
  "domain": "Complicated",
  "domain_confidence": 0.89,
  "domain_entropy": 0.34,
  "guardian_verdict": "approved",
  "response": "Analysis summary...",
  "requires_human": false,
  "reasoning_chain": [
    {"node": "router", "action": "Classified", "confidence": "high"},
    {"node": "causal_analyst", "action": "Estimated effect", "confidence": "medium"}
  ],
  "causal_result": {
    "effect": 0.15,
    "unit": "percentage points",
    "ci_low": 0.08,
    "ci_high": 0.22,
    "refutations_passed": 3,
    "refutations_total": 3
  },
  "bayesian_result": null,
  "guardian_result": {
    "verdict": "approved",
    "policies_passed": 5,
    "policies_total": 5,
    "risk_level": "low"
  }
}

Domain Routing Logic

Domain	Confidence	Entropy	Agent
Clear	> 0.95	< 0.2	`deterministic_runner`
Complicated	> 0.85	< 0.5	`causal_analyst`
Complex	> 0.7	0.5-0.8	`bayesian_explorer`
Chaotic	Any	> 0.9	`circuit_breaker`
Disorder	< 0.85	Any	`human_escalation`

Python API

For direct invocation:

python

from src.workflows.graph import run_carf

result = await run_carf(
    user_input="Why did costs increase 15%?",
    context={"domain": "finance"}
)

print(f"Domain: {result.cynefin_domain}")
print(f"Response: {result.final_response}")
print(f"Causal Evidence: {result.causal_evidence}")

Troubleshooting

Query Routed to Disorder

•Low confidence from router (< 0.85)
•Ambiguous or vague query
•Solution: Rephrase with more specificity

Guardian Rejected Action

•Policy violation detected
•Check guardian_result.violations in response
•May require human approval via HumanLayer

Timeout on Complex Queries

•PyMC inference can be slow
•Increase timeout or reduce sample count
•Check for large dataset sizes