escalation-governance

Escalate only after investigation reveals genuine complexity, high-stakes decisions, novel patterns, or ambiguity requiring deeper reasoning. First understand the problem, justify the need, define scope and success criteria, then move to a capable model.

Investigation-first means analyzing the task before escalating. Require problem understanding, clear justification, defined scope, success criteria, and a safe handoff protocol back to faster models once the complex portion is resolved.

Escalation governance provides triggers and decision frameworks for this. Use it to identify when tasks involve genuine complexity, high-stakes outcomes, or novel patterns where deeper reasoning justifies a more capable model over faster alternatives.

Unnecessary escalation wastes resources and latency on tasks that faster models handle adequately. Governance enforces criteria—complexity, stakes, novelty, ambiguity—to ensure escalation occurs only when the reasoning depth of a capable model is genuinely required.

Yes. Escalation governance includes protocol for safe handoff back to faster models once the complex reasoning portion is complete, ensuring you use the right model tier for each phase of the task.