Classification

Evidentiary Hierarchy of Transition Checks

StateM categorizes transition verification checks into several distinct tiers of evidentiary strength:

  • Command and Predicate Checks: Deterministically evaluated by the host environment (e.g., exit codes or file presence), offering independently reproducible verification.
  • Manual Checks: Mandate an explicit decision or sign-off by a human operator.
  • Checklist and Message Checks: Structured self-attestations requiring the agent to confirm obligations, which do not constitute independent verification.
  • LLM-Review Checks: Incorporate semantic judgment via an external model call, providing flexible heuristics without deterministic guarantees.

This hierarchy ensures structured attestations or model self-reports are not conflated with deterministic host-evaluated evidence.

0

1

Updated 2026-09-11

Tags

Prep Sessions

Engineering State-Bound Execution Runtimes for Autonomous Agents @ University of Michigan - Ann Arbor

Ch.2 Agent Runtime Design and Execution Framework - Engineering State-Bound Execution Runtimes for Autonomous Agents @ University of Michigan - Ann Arbor

StateM Agent-Native Runtime Architecture - Engineering State-Bound Execution Runtimes for Autonomous Agents @ University of Michigan - Ann Arbor

Context and Contract Boundaries in Phase Execution - Engineering State-Bound Execution Runtimes for Autonomous Agents @ University of Michigan - Ann Arbor

Related