test-authority

Escalate when the task presents genuine complexity or reasoning depth that exceeds current capability—novel patterns, high-stakes decisions, or knowledge gaps requiring deeper analysis. Escalation is justified when the problem itself demands it, not due to thrashing, time pressure, or unfounded safety concerns.

A knowledge gap means the model lacks specific information but has sufficient reasoning capacity to process it if provided. A capability gap means the task requires reasoning depth, pattern recognition, or judgment beyond the current model's architecture. Escalate for capability gaps; investigate or provide context for knowledge gaps.

Escalation decisions require a formal framework with documented criteria: define what constitutes genuine complexity, establish decision checkpoints where escalation is evaluated, exclude false triggers like performance pressure or unvalidated safety flags, and ensure all escalations are traceable and justified for audit and transparency.

Yes. Use a pressure-test scenario with defined decision points to evaluate when escalation is warranted versus unnecessary. This lets you validate your escalation protocol, identify judgment gaps, and refine authority thresholds in a controlled environment before deploying to live workflows.

Document the specific condition triggering escalation, the reasoning framework applied, which criteria were evaluated, and the outcome. Implement agent-schema guidance that captures decision metadata, ensuring every escalation has an audit trail linking the trigger, decision logic, and result for compliance and learning.