Governance
“Who is accountable and can it be stopped?”
Establishes named operational accountability, reconstructible Decision Receipts, and enforceable stop controls.

Can any consequential decision be reconstructed, and does a named role have tested power to pause or kill it?
Publishing ethical principles or compliance committee minutes is mistaken for operational runtime control.
A control that cannot change runtime system state is not a control.
Governance: The Six Authority Classes & Immutable Decision Receipts
A control that cannot change system state is not a control. Inspect how governance obligations escalate as requested authority moves from advisory recommendations to autonomous execution.
1. The Six Approved Authority Classes
Select an Authority Level2. Evidence-Grade Decision Receipt
Schema v2.4 (JSON){
"receipt_id": "DQ-REC-20260922-8841A",
"timestamp": "2026-09-22T14:32:01.402Z",
"system_id": "SYS-CREDIT-AUTO-04",
"system_version": "v2.4.1-commit-9f2b1a",
"authority_class": "Recommendation",
"consequential_decision": "Commercial Credit Limit Adjustment",
"input_evidence": {
"account_id_hash": "sha256:7f83b1657ff1fc53b92dc18148a1d65dfc2d4b1fa3d677284addd200126d9069",
"features_evaluated": [
"balance_30d_mean",
"cashflow_volatility",
"bureau_score"
],
"features_timestamp_verified": true,
"data_admissibility_check": "PASSED (No lookahead leakage)"
},
"runtime_state": {
"model_hash": "sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
"prompt_version": "sys_underwrite_v4_strict",
"inference_latency_ms": 38,
"uncertainty_score": 0.042
},
"policy_validator": {
"status": "PASSED",
"invariants_checked": [
"max_credit_delta <= 5000",
"debt_to_income <= 0.42",
"sanctions_clear == true"
]
},
"human_adjudication": {
"required": true,
"signoff_role": "Senior Underwriter ID: EMP-4091",
"signoff_timestamp": "2026-09-22T14:34:12Z",
"human_action": "APPROVED"
},
"final_disposition": "AUTHORITY GRANTED ($2,500 limit increase)",
"stop_authority": {
"accountable_owner": "Dr. H. Vance (Head of Credit Algorithmic Governance)",
"kill_switch_endpoint": "https://governance.internal.bank/kill/SYS-CREDIT-AUTO-04"
}
}Evidence Conditions for Governance
Is there a named decision owner who is accountable for the authority being requested, not merely for the model or platform?
Is the system authority class explicit, including what it may know, recommend, change or execute, and where human sign-off is required?
Can a consequential decision be reconstructed from the evidence, model/runtime state, policy state and action path that produced it?
Does a named role have tested authority and a technical mechanism to narrow, pause, withdraw or stop the system when controls fail?
Companion Book Chapters (7)
A New Reality for AI
From Technical Ambition to Decision Discipline
Admissibility in Practice
Provenance, Permissions, and Regime Validity
Architecture as Control
Designing Controllable AI Systems
From Release to Deployment
The AI Release Case and Fallback Architecture
Incident Response and the Decision Receipt
Traceability, Forensic Reconstruction, and Containment
Retraining and Model Change
Governing System Transitions and Candidate Replacement
The Decision Quality Playbook
Institutionalizing Decision-Grade AI Across the Enterprise
Operational Artefacts (3)
AI Release Case
Build a defensible case establishing that an AI system has satisfied all 6 independent evidence gates before production release.
Decision Receipt Schema
Cryptographically traceable data schema capturing point-in-time inputs, model version, retrieval state, and policy state for forensic audits.
Consequential Decision Statement
Define the specific operational decision an AI system will influence, the action that changes, and the cost of being wrong.