Handbook

Task `llm_boolean_gate` v1

Given facts (object) and a yes/no question, return yes, numeric confidence, and a short reason.

Updated

Task llm_boolean_gate v1

Summary

Given facts (object) and a yes/no question, return yes, numeric confidence, and a short reason.

Inputs

Field Type Required Notes
facts object yes Structured facts for the model (may be {}).
question string yes Closed yes/no question.

Output

{
  "yes": false,
  "confidence": 0.8,
  "reason": "Facts do not state an explicit deadline."
}

Examples

Minimal:

{ "facts": { "x": 1 }, "question": "Is x positive?" }

Failure modes

  • SchemaFailure — missing yes / confidence / reason or wrong types after parse.
  • ParseFailure — assistant output not a JSON object.
  • GatewayFailure — transport/provider errors.

Verification

JSON schema validation against contract.json output_schema after parse.

Changelog

  • v1 — Initial boolean gate.