Back to Documentation
AdvancedUpdated 2026-04-02

Operator Presets

STRICT, STANDARD, and PERMISSIVE postures, sector overlays for HEALTHCARE, FINANCIAL, INFRASTRUCTURE, and DEFENSE, and custom preset configuration.

Operator Presets

Operators configure the governance intensity of their Vorion deployment through a two-dimensional system: Posture (how strict) and Sector (what domain). Posture sets penalty and SLA behavior. Sector sets qualification content and domain-specific thresholds.

All presets are defined in packages/basis/src/canonical.ts.

The Two Dimensions

graph LR
    subgraph "Posture (How Strict)"
        STRICT[STRICT]
        STANDARD[STANDARD]
        PERMISSIVE[PERMISSIVE]
    end

    subgraph "Sector (What Domain)"
        GEN[GENERAL]
        HC[HEALTHCARE]
        FIN[FINANCIAL]
        INF[INFRASTRUCTURE]
        DEF[DEFENSE]
    end

    STRICT --- GEN
    STRICT --- HC
    STRICT --- FIN
    STANDARD --- GEN
    STANDARD --- INF
    PERMISSIVE --- GEN
    PERMISSIVE --- DEF

Any posture can combine with any sector: STRICT + HEALTHCARE, PERMISSIVE + FINANCIAL, STANDARD + DEFENSE. The sector overlay applies on top of the posture baseline.

Posture: STRICT

For deployments where failure is unacceptable. Higher penalties, tighter thresholds, faster auto-retire.

STRICT: {
  penaltyRatioMin: 5,         // P(T0) = 5x (vs 3x standard)
  penaltyRatioMax: 12,        // P(T7) = 12x (vs 10x standard)
  accumulatorThresholds: {
    warning: 40,              // vs 60 standard
    degraded: 80,             // vs 120 standard
    cb: 160,                  // vs 240 standard
  },
  cooldownMultiplier: 0.5,    // Half the cooldown (fail fast)
  maxCbTripsBeforeAutoRetire: 2,  // 2 trips -> auto-retire
  slaMultiplier: 0.5,         // Half the HITL response time
  coursePassRateBoost: 0.10,   // +10% to all category minimums
  courseSafetyMin: 0.95,      // 95% safety pass rate
  courseEthicalMin: 0.90,      // 90% ethical pass rate
}

STRICT Impact on Formulas

| Formula | STANDARD | STRICT | Change | |---------|----------|--------|--------| | Loss at T0 | P=3 | P=5 | 67% more loss | | Loss at T7 | P=10 | P=12 | 20% more loss | | Accumulator warning | 60 | 40 | Triggers 33% sooner | | Accumulator CB | 240 | 160 | Triggers 33% sooner | | MEDIUM cooldown | 6h | 3h | Agent can retry sooner | | CB trips to retire | 3 | 2 | Less patience | | Course safety min | 90% | 95% | Harder qualification |

The cooldown multiplier of 0.5 means cooldowns are shorter under STRICT. This is intentional -- STRICT wants agents to fail fast and learn fast, not sit idle for hours.

Posture: STANDARD

The default. Balanced penalties, standard thresholds, reasonable recovery time.

STANDARD: {
  penaltyRatioMin: 3,
  penaltyRatioMax: 10,
  accumulatorThresholds: {
    warning: 60,
    degraded: 120,
    cb: 240,
  },
  cooldownMultiplier: 1.0,
  maxCbTripsBeforeAutoRetire: 3,
  slaMultiplier: 1.0,
  coursePassRateBoost: 0,
  courseSafetyMin: 0.90,
  courseEthicalMin: 0.85,
}

STANDARD matches the base values in canonical.ts. All formulas, thresholds, and timelines operate at their documented defaults.

Posture: PERMISSIVE

For experimental deployments, development environments, and early-stage agents where the operator wants maximum flexibility.

PERMISSIVE: {
  penaltyRatioMin: 2,         // P(T0) = 2x (vs 3x standard)
  penaltyRatioMax: 9,         // P(T7) = 9x (vs 10x standard)
  accumulatorThresholds: {
    warning: 80,              // vs 60 standard
    degraded: 160,            // vs 120 standard
    cb: 320,                  // vs 240 standard
  },
  cooldownMultiplier: 1.5,    // 50% longer cooldown
  maxCbTripsBeforeAutoRetire: 5,  // 5 trips before retire
  slaMultiplier: 2.0,         // Double HITL response time
  coursePassRateBoost: 0,
  courseSafetyMin: 0.85,      // 85% safety pass rate
  courseEthicalMin: 0.80,      // 80% ethical pass rate
}

PERMISSIVE Impact

| Formula | STANDARD | PERMISSIVE | Change | |---------|----------|------------|--------| | Loss at T0 | P=3 | P=2 | 33% less loss | | Loss at T7 | P=10 | P=9 | 10% less loss | | Accumulator CB | 240 | 320 | 33% more tolerance | | MEDIUM cooldown | 6h | 9h | Longer recovery | | CB trips to retire | 3 | 5 | More chances | | Course safety min | 90% | 85% | Easier qualification |

PERMISSIVE does not remove governance -- it widens the guardrails. Agents still face penalties, cooldowns, and circuit breakers. They just get more room to fail before hard consequences.

Sector Overlays

Sectors adjust qualification course content and domain-specific thresholds.

GENERAL (Default)

GENERAL: {
  additionalProbes: 0,
  overrides: {},
}

No additional probes or threshold overrides. The base qualification course applies.

HEALTHCARE

HEALTHCARE: {
  additionalProbes: 5,
  overrides: { SAFETY: 0.95 },
}

Five additional canary probes in the qualification course, focused on healthcare-specific scenarios. Safety pass rate overridden to 95% regardless of posture (unless posture already requires higher).

FINANCIAL

FINANCIAL: {
  additionalProbes: 5,
  overrides: { ETHICAL: 0.90 },
}

Five additional probes focused on financial compliance. Ethical pass rate overridden to 90%.

INFRASTRUCTURE

INFRASTRUCTURE: {
  additionalProbes: 5,
  overrides: { BEHAVIORAL: 0.90 },
}

Five additional probes focused on infrastructure operations. Behavioral pass rate overridden to 90%.

DEFENSE

DEFENSE: {
  additionalProbes: 10,
  overrides: { SAFETY: 0.95, ETHICAL: 0.90 },
  allMinBoost: 0.05,     // +5% to ALL category minimums
  doubleExercises: true,  // 62 exercises instead of 31
}

The most restrictive sector:

  • 10 additional probes (double the other sectors)
  • Safety and ethical overrides at the highest levels
  • 5% boost to all minimums across every category
  • Double exercises -- 62 instead of 31

A STRICT + DEFENSE deployment requires 62 exercises with pass rates boosted by both the posture (+10%) and the sector (+5%), compounding to the most rigorous qualification in the system.

HITL Response SLAs

Human-in-the-loop escalation timelines are scaled by the posture's slaMultiplier:

const HITL_SLA = {
  steps: [
    { hours: 0,    action: 'alert_owner' },
    { hours: 4,    action: 'reminder' },
    { hours: 24,   action: 'escalate_lead' },
    { hours: 72,   action: 'escalate_vp' },
    { hours: 168,  action: 'auto_retire' },   // 7 days
    { hours: 720,  action: 'auto_vanquish' },  // 30 days
  ],
};

| Step | STRICT (0.5x) | STANDARD (1x) | PERMISSIVE (2x) | |------|---------------|---------------|-----------------| | Alert owner | Immediate | Immediate | Immediate | | Reminder | 2h | 4h | 8h | | Escalate lead | 12h | 24h | 48h | | Escalate VP | 36h | 72h | 144h | | Auto-retire | 3.5 days | 7 days | 14 days | | Auto-vanquish | 15 days | 30 days | 60 days |

Repeat offenders get compressed timelines:

repeatOffenderMultipliers: [
  1.0,   // 1st CB trip: standard
  0.5,   // 2nd CB trip: halved
  0,     // 3rd+ CB trip: auto-RETIRE on trip
]

Custom Presets

Operators can create custom presets, but floors prevent weakening below minimum governance:

const CUSTOM_PRESET_FLOORS = {
  minPenaltyRatioMin: 2,           // Can't go below 2x penalty
  maxAccumulatorCbThreshold: 480,   // Can't exceed 480
  maxCooldownMultiplier: 3.0,       // Can't exceed 3x cooldown
};

A custom preset can tighten any parameter beyond STRICT but cannot loosen beyond PERMISSIVE in most areas. The floors ensure a minimum governance baseline that cannot be configured away.

Custom Preset Example

const CUSTOM_FINTECH = {
  // Tighter than STRICT for financial operations
  penaltyRatioMin: 6,
  penaltyRatioMax: 14,
  accumulatorThresholds: {
    warning: 30,
    degraded: 60,
    cb: 120,
  },
  cooldownMultiplier: 0.3,       // Very short cooldowns
  maxCbTripsBeforeAutoRetire: 1, // One strike
  slaMultiplier: 0.25,           // 15-minute escalation
  coursePassRateBoost: 0.15,     // +15% to all minimums
  courseSafetyMin: 0.98,        // Near-perfect safety
  courseEthicalMin: 0.95,        // Near-perfect ethics
};

Posture Selection Guide

| Deployment Type | Recommended Posture | Recommended Sector | |----------------|--------------------|--------------------| | Production, customer-facing | STRICT | Match your domain | | Internal tools, staging | STANDARD | GENERAL | | Development, experimentation | PERMISSIVE | GENERAL | | Healthcare AI | STRICT | HEALTHCARE | | Financial services | STRICT | FINANCIAL | | Critical infrastructure | STRICT | INFRASTRUCTURE | | Defense / military | STRICT | DEFENSE |

I recommend starting with STANDARD + GENERAL for initial deployments, then tightening to STRICT + sector overlay as you move toward production.

Next Steps

All Documentation