A guardrail returns structured JSON output for every record. This output is consistent across all policy types and cannot be modified. At minimum it includes the decision: flagged (true/false).
Depending on policy mode, it may also include:
-
Moderation: triggered categories/reasons
-
PII: detected entity types/segments (and any optional masked representation if returned)
-
Custom policy: short explanation
Check the value of this structured output for deterministic routing and mapping in downstream flow steps. You do not need to parse raw model output. The structured result is ready for mapping in Mapper/Transformation 2.0 or standard step mappings.