Governance and policy

Proposal filtering, scoring, hysteresis, conflict handling, and policy explainability.

Governance pipeline

1. proposal generation
2. goal/lane filtering
3. conflict and cooldown checks
4. policy and trust checks
5. cost/risk scoring
6. execution with audit record

Effective score

effectiveScore = utility − (costWeight × trustedCost) − (riskWeight × trustedRisk)

Vitruvian also applies minimum cost/risk floors by side-effect level so modules cannot under-report risk.

Hysteresis

Near-tie winners can stay selected to avoid oscillation between equivalent modules.

Policy commands

• policy validate <policyFile>
• policy explain <request>

When no custom policy is set, EnterpriseSafe defaults are used:

• read-only intent: allowed
• write/destructive intent: approval required

Governance gotchas

• If your module never wins, tune utility and metadata before tuning risk floors.
• Cooldowns can make a healthy module appear unavailable if TTL is too aggressive.
• Conflicts are explicit; declare them to avoid contradictory execution paths.