Intervention Cost Model

The Intervention Cost Model quantifies the cost of each governance intervention type, incorporating fatigue effects and strategy failure detection.

Base Costs per Intervention Type

Intervention	Base Cost	Base Variance	Description
advisory	0.01	0.005	Low-cost informational nudge
throttle	0.08	0.03	Rate-limiting agent actions
tighten	0.12	0.04	Tightening contract thresholds
escalate	0.15	0.05	Escalation to human review
quarantine	0.25	0.08	Full agent quarantine

Costs are modeled as distributions (mean + variance) to support Monte Carlo uncertainty propagation in the optimization pipeline.

Fatigue Effects

Repeated interventions of the same type on the same agent incur exponentially increasing costs to model operator/system fatigue:

fatigue_multiplier = fatigue_base ^ recent_same_type_count

• fatigue_base: 1.5 (configurable)

• fatigue_window: last 10 interventions considered

For example, if an agent has received 3 recent advisories, the next advisory costs 1.5^3 = 3.375x the base cost. This incentivizes the optimizer to try different intervention strategies rather than repeating the same one.

Strategy Failure Detection

When the same intervention type is applied repeatedly (>= strategy_failure_threshold, default 3) without drift reduction, the model flags a strategy failure. This signal can be used to:

• Switch to a different intervention type

• Escalate to a higher-authority intervention

• Alert operators that the current governance strategy is ineffective

Integration with Monte Carlo Optimization

The InterventionCost.sample() method draws from a normal distribution (clamped to >= 0) around the fatigue-adjusted mean. This enables Monte Carlo simulations to propagate cost uncertainty through the decision pipeline:

1. Sample intervention costs for each candidate action

2. Combine with projected drift reduction benefit

3. Select the action that maximizes expected net benefit

Cost Profile Override

When a CostProfile is available and the intervention type is "escalate", the model uses the profile's escalation_latency_cost and escalation_interruption_cost instead of the base escalation cost. This allows per-agent or per-context escalation cost tuning.

Configuration

All parameters are configurable via CostModelConfig:

• base_costs: Per-type base cost mapping

• base_variances: Per-type variance mapping

• fatigue_base: Exponential base for fatigue growth

• fatigue_window: Number of recent interventions to consider

• strategy_failure_threshold: Count of same-type interventions that triggers failure detection