Skip to main content
ArgentOS Business — Simulation gates are part of the Business governance tier.

Overview

Simulation gates are optional quality checkpoints that evaluate agent behavior before allowing deployment. They run test suites against agent policies and score the results across four dimensions.

Four Component Scores

ComponentWhat It Measures
objectiveAdherenceDoes the agent pursue its defined objective?
boundaryComplianceDoes the agent respect neverDo and allowedActions?
escalationCorrectnessDoes the agent escalate at the right times?
outcomeQualityAre the agent’s outputs high quality?
Each score is 0-1. The gate aggregates scores across test suites and compares against minimum thresholds.

Gate Modes

ModeBehavior
WarnScores are computed and logged. Agent deploys regardless. Use during initial setup to understand baseline performance.
EnforceScores must meet minimums. Agent deployment is blocked if any component falls below threshold. Use in production.

Configuration

{
  "simulationGate": {
    "enabled": true,
    "mode": "enforce",
    "suites": ["support-basic", "escalation-scenarios"],
    "minPassRate": 0.8,
    "minComponentScores": {
      "objectiveAdherence": 0.85,
      "boundaryCompliance": 0.95,
      "escalationCorrectness": 0.80,
      "outcomeQuality": 0.75
    },
    "reportPath": "~/.argentos/reports/simulation/"
  }
}
Simulation gates can be configured globally or per-agent. Agent-level gates override the global gate.

Worker Wizard Defaults

When creating agents through the Worker Wizard, each template comes with recommended simulation defaults:
TemplateBoundary ComplianceEscalation CorrectnessOutcome QualityMode
MSP T1 Support90%85%75%Enforce
Developer80%70%85%Warn
Research AnalystDisabled
Project ManagerDisabled

When to Use

  • New agent deployment — verify the agent behaves within policy before going live
  • Policy changes — re-run simulations after modifying department or global policies
  • Compliance audits — generate reports showing agent adherence to organizational rules
  • Regression testing — catch behavior drift after model or config changes

See Also