The Three-Body Problem of Classified AI - Practitioner Edition: Operationalizing Trust-Governed AI on the High Side (The Harness Doctrine)

ou will not authorize a classified AI system from a whiteboard. You will authorize it from artifacts.

The Foundations Edition diagnosed the problem. The Implementation Edition built the architecture. This is the volume you take to the work.

The Practitioner's Edition compresses the theory into three recap chapters and turns the rest into forty-five worksheets, checklists, scorecards, and configuration templates you can fill in by hand or wire into your tooling. System Assessment Questionnaire. Authorization Threshold Checklist. Delegation Token Design Worksheet. RAG Architecture Configuration Template. Policy-as-Code Template Library. Evidence Stream Implementation Checklist. ATO Pre-Submission Readiness Scorecard. Governance Gate Definition. Runtime Detection Configuration. Prompt Injection Mitigation Checklist. Pre-Flight Checklist. Incident Response Playbook. A1-A7 Adversary Response Matrix. Authorization Package Assembly Checklist.

Open to any worksheet and you get the same three things: what the property requires, the evidence the assessor will demand, and a line to write your answer on. No re-derivation. No reread. Every tool maps back to a P1-P8 property and an evidence tier, so nothing you fill in is decorative.

The appendices carry the sample configurations, the testing scripts, the Quick Reference, and a Quick Start Guide that gets a minimal viable harness running before the next standup.

If the control ran, evidence exists. If evidence does not exist, the control never ran.

Written for security engineers, DevSecOps teams, system operators, and implementation teams with a deadline and an authorization package to assemble. Keep this volume on the desk, not the shelf.