Papers
Formal papers and long-form research.
- Paper
Postmortem Before the Fact
Backtesting AI-Assisted Failure Forecasting for Strategic Decision-Making
Language models make premortems cheap to repeat, but a plausible failure narrative is not a validated forecast. This paper proposes Failure-First Forecasting, a protocol that freezes a plan, generates causal failure pathways, remediates them, reforecasts the revised plan, and scores the whole process against what actually happened, with temporal isolation treated as a first-order requirement.
- Paper
From Inference to Admission
A Decision-Theoretic Specification for Governing AI-Derived Operational State
A decision-theoretic specification for governing the boundary at which AI-derived values become authoritative operational state: dependency-group admission control, non-additive loss, commit/review/abstain under finite review capacity, and an evaluation contract that reports consistency rather than accuracy.
- Paper
From Copilots to Controlled Digital Operations
A Practical Architecture for Governed Multi-Agent Orchestration and Stateful Workflow Graphs
An architecture for inserting non-deterministic agent reasoning into durable enterprise workflows without surrendering control: evidence and trust as first-class workflow state, separation of reasoning authority from execution authority, routing on evidence and deterministic verification rather than model confidence, and a fit test for when agents should not be used.