Read a workflow result

app.run(...) returns one ExecutionResult. Read execution.status first, then use payload for the business value and flows for detail. Do not infer success from the presence of a payload: on failure it preserves the accepted input rather than fabricating a successful output.

Field	What it tells your application
payload	The workflow's configured output projection; without one, the accepted input payload.
metadata	Accepted caller context.
flows	Records by flow instance ID, with local steps and optional projected result.
transitions	Routes or terminal outcomes selected at flow boundaries.
execution	Run ID, workflow, revision, status, measured usage, and safe error on failure.

A completed step result is at result.flows[flow_id].steps[step_id].result. Its result can be JSON null; an omitted result means something else. Flow and step records can be completed, needs_review, failed, cancelled, or skipped. The root execution cannot be skipped.

Read a decision

A decision step with one configured question exposes one result object. A step with questions exposes {"results": [...]} in question order. A single choice can look like this:

{
  "questionId": "request_kind",
  "type": "choice",
  "answerability": {"status": "answerable", "issues": []},
  "answer": {"optionId": "statement"},
  "reason": "The message asks for a statement.",
  "evidence_strength": "strong"
}

Your code should branch on answerability.status, stable issues, and the typed answer. reason explains the assessment to a person; its wording is not a route key. evidence_strength describes support for the whole reported assessment, including answerability and any abstention. strong can therefore describe a well-supported not_answerable result. limited is a weaker but permissible interpretation; null means strength was not assessed. These are not probabilities, confidence scores, or automatic action thresholds.

answerable has a substantive answer. partially_answerable is available for supported subsets of multiselect and request_units. not_answerable and undetermined have issue codes and no substantive answer, except that a predicate uses {"value": "unknown"}. For a singular choice, selection.origin: fallback may record an authored fallback category while the native answer remains null and the step remains needs_review.

Foliqant validates the structure and correspondence of model output. It does not prove the explanation or strength rating is correct. Use reviewed evaluation cases for that question. The exact question and answer shapes are in the reference; see decision policy for criteria and fallback design.