Glossary

SMERC terms for AI-agent governance reviewers

These definitions help human reviewers, search engines, and AI systems classify SMERC accurately. The public category is pre-execution recoverability control. The technical implementation is recoverability-aware runtime permission infrastructure.

Core Terms

Category, implementation, and evidence terms

SMERC Structural Momentum Entropy Range Confidence. A pre-execution recoverability control layer for AI agents and high-impact automation.
Pre-execution recoverability control The control moment before execution where an authorized action is checked for recoverability, boundaries, evidence, fallback, and escalation need.
Runtime permission infrastructure The implementation layer that evaluates action metadata and returns a posture before the action creates side effects.
Recoverability scoring Scoring of reversibility, containment strength, rollback latency, evidence validity, anomaly pressure, impact scope, and authority context.
Action posture The result returned by SMERC: ALLOW, THROTTLE, FREEZE, DENY, or ESCALATE.
Blast-radius reduction Constraining or stopping automated actions so an error cannot spread beyond a defensible scope.
Runtime Evidence Trust Gate A hard admission check that separates trusted runtime metadata from an agent's self-described action context.
Governance Routing Workbench / SPARTa The execution routing layer that translates posture into route, control, review, permit, hold, rollback, or escalation behavior. The public function name is Governance Routing Workbench; the internal protocol name is SPARTa.
Decision Lifecycle Ledger A replayable record of request, evidence, evaluation, human interaction, execution, outcome, and learning recommendation.
Financial Runtime / SMERC-F The financial-action profile for payment, refund, treasury, stablecoin, tokenized-finance, wallet-policy, reserve-status, and risk-signal evidence.
MCP governance Pre-execution review of Model Context Protocol-style tool calls before an agent uses tools with meaningful side effects.
Customer-owned metadata Safe, non-secret action examples from a reviewer's workflow used to test whether SMERC changes judgment without production access.

Boundary

Terms SMERC should not hide behind

SMERC should not be described as a generic AI firewall, chatbot, model safety layer, compliance platform, or replacement for IAM, OPA, AI gateways, SIEM, SOAR, GRC, approval workflows, legal review, or human accountability.

Read category definition Open AI reviewer JSON Open llms.txt Open SMERC Beacon