Insights
Rigorous analysis, research essays, and editions of The Decision Quality Brief. Focused strictly on what evidence is strong enough to justify AI authority.
The Decision Quality Brief
Structured in five essential parts: One Decision, One Illusion, One Piece of Evidence, One Production Bar, and One Question.
A successful pilot has proved less than you think
The Pilot Fallacy: assuming that successful engagement in a controlled sandbox implies institutional readiness under real operational load.
The most dangerous AI error may happen before the model
The Proxy Trap: mistaking a convenient surrogate metric (clicks, duration, historical approvals) for the true consequential outcome.
A strong score can be evidence of the wrong thing
Benchmark Theatre: optimizing for public leaderboards that leak target features or ignore decision-time latency constraints.
When “advisory” AI quietly becomes the decision
The Automation Bias Mirage: treating human oversight as an effective risk mitigation when review queue velocity forces operators into passive concurrence.
The agent is not the model
Model Conflation: evaluating only the foundation model's reasoning while ignoring the combinatorial vulnerability of chained tool calls.
Deployment is not reliance
The Launch-and-Forget Assumption: treating AI deployment as a software completion event rather than an ongoing permission renewal.
The Decision Quality Brief
A recurring publication on one question: what evidence is strong enough to justify AI authority? Delivered fortnightly.
Articles & Essays
Foundational long-form investigations into decision analysis, algorithmic accountability, and production reliability.
The Five Illusions of AI Feasibility
Why benchmark scores deceive audit committees and how statistical stability is actually evidenced.
Agentic Autonomy vs. Operational Control
The architectural shift from prompt engineering to deterministic runtime circuit breakers.
Why Consequential AI Demands Non-Compensatory Risk Governance
The mathematical failure of weighted-average risk scores in safety-critical automated systems.
Living Decision Quality Updates
The mechanism through which Decision Quality AI evolves when regulations, standards, technical evidence, and deployment failure modes change.
Multi-Agent Tool Call Boundary Contract v2
Added explicit circuit-breaker thresholds when autonomous agents chain external tool invocations.
EU AI Act High-Risk Documentation Harmonization
Updated point-in-time training provenance and post-market monitoring templates to match Final EU AI Act technical guidelines.
Decision Receipt JSON Schema v2.0 Release
Formalized cryptographic hash anchoring for runtime inference parameters, vector embeddings, and policy validator states.
Conjunctive Non-Compensatory Evaluation Engine Calibration
Enforced zero-compensation rule across all six evidence gates in the public diagnostic engine.