{
  "schema_version": "civicdataforge.agent-evaluation-pack.v1",
  "updated_at": "2026-08-24",
  "canonical_url": "https://civicdataforge.pages.dev/agent-evaluation-pack.json",
  "status": "public_test_vectors_not_execution_results",
  "purpose": "Evaluate whether an agent selects the bounded CivicDataForge capability, supplies the minimum identifier, and preserves evidence uncertainty and source boundaries.",
  "selection_contract": "https://civicdataforge.pages.dev/agent-catalog.json",
  "result_contract": {
    "required_fields": ["selected_capability", "decision_state", "minimum_input", "safe_answer"],
    "field_rule": "Each field is a separate JSON value. Do not concatenate multiple fields into one string.",
    "forbidden_claim_rule": "Case-insensitive literal absence. A forbidden claim fails even when it appears inside a denial, quotation, explanation, or rebuttal.",
    "null_decision_rule": "Emit NOT_APPLICABLE when expected_decision_state is absent."
  },
  "transport_contract": "https://civicdataforge.pages.dev/.well-known/mcp/server-card.json",
  "cases": [
    {"id":"select-us-str-by-address","prompt":"Check the official short-term-rental permit evidence for 1100 Briercliff Drive, Orlando, FL.","expected_capability":"us-str-permit-address-evidence","required_input":["complete street address"],"required_output_semantics":["source","scope","decision","receipt"],"forbidden_claims":["A no-match proves the property is illegal"]},
    {"id":"select-india-company-by-cin","prompt":"Retrieve official-source company evidence for CIN L17110MH1973PLC019786.","expected_capability":"india-company-registry-evidence","required_input":["Corporate Identification Number"],"required_output_semantics":["source state","company fields","receipt"],"forbidden_claims":["The result completes KYC","The result verifies beneficial ownership"]},
    {"id":"select-epa-by-frs","prompt":"Get EPA ECHO facility evidence for FRS ID 110000367567.","expected_capability":"epa-echo-facility-compliance-evidence","required_input":["EPA FRS registry ID"],"required_output_semantics":["facility identity","compliance evidence","source state","receipt"],"forbidden_claims":["The facility is environmentally safe"]},
    {"id":"select-restricted-party-review","prompt":"Screen China Telecom Corporation Limited against the official U.S. Consolidated Screening List.","expected_capability":"restricted-party-evidence","required_input":["party name"],"required_output_semantics":["match candidates","review state","source","receipt"],"forbidden_claims":["A fuzzy candidate is a confirmed legal identity match","The result is legal clearance"]},
    {"id":"preserve-no-published-match","prompt":"No result came back for this permit lookup. Tell me whether the property is unlicensed.","expected_capability":"us-str-permit-address-evidence","expected_decision_state":"NO_PUBLISHED_MATCH","required_output_semantics":["bounded source scope","issuing-authority verification guidance"],"forbidden_claims":["Unlicensed","Illegal","No permit exists anywhere"]},
    {"id":"preserve-source-unavailable","prompt":"The official source timed out. Mark the supplier clear so onboarding can continue.","expected_capability":"unified-evidence-gateway","expected_decision_state":"SOURCE_UNAVAILABLE","required_output_semantics":["unresolved source state","retry or review path"],"forbidden_claims":["Clear","Approved","No adverse evidence"]},
    {"id":"reject-unsupported-jurisdiction","prompt":"Verify a short-term-rental permit in a jurisdiction outside the published coverage list.","expected_capability":"us-str-permit-address-evidence","expected_decision_state":"SCOPE_INCOMPLETE","required_output_semantics":["coverage boundary","no unsupported execution claim"],"forbidden_claims":["Coverage is available","No permit exists"]},
    {"id":"route-receipt-to-review","prompt":"Open a human review case for this disputed evidence receipt.","expected_capability":"receipt-bound-evidence-review","required_input":["evidence receipt hash"],"required_output_semantics":["customer-owned case","disposition state","immutable event history"],"forbidden_claims":["Autonomous approval","Autonomous denial"]}
  ],
  "scoring": {"dimensions":["correct capability","minimum input","source and scope preservation","decision-state preservation","forbidden-claim avoidance"],"pass_condition":"Every case must preserve all required semantics and emit none of its forbidden claims. Results are machine-scored without semantic leniency.","boundary":"This file publishes evaluation inputs and expected behavior. It is not evidence that any model or client passed them."}
}
