Notes on software, systems, and the decisions behind them. Questions I’m exploring as I build.
Ideas in progress, open questions, and the thinking behind the systems.
Editorial notes, not published studies.
A useful decision system preserves the evidence, the rule version, and the person responsible for the outcome. The question is not only whether a result can be reproduced, but whether someone can challenge its reasoning.
Open question: What should a system retain so that a decision remains understandable six months later?
Automation should have a clearly defined envelope of authority. Outside it, the system should stop, explain what is missing, and return control to a person. A refusal to act can be a successful outcome.
Open question: Which actions are reversible, and which require explicit human approval?
Every boundary is an opportunity to make trust explicit. Validation checks shape; authorization checks permission; audit records explain what happened. None of these replaces the others.
Open question: Where does untrusted input first acquire the authority to change state?
A question, an idea, or just a hello.
Start a conversation