Evidence before confidence
A confident answer is still a claim. The useful question is what supports it, what limits it and what remains unknown.
Writing
Notes on evidence, control and responsibility.
Writing is where the doctrine is tested as an argument: against sources, failure modes, organisational reality and the consequences of action.
No text appears here as a publication until it has an author, a version, publication and review dates, sources, an AI-use disclosure and a corrections path.
Designing for fallibility
Models, data, tools and people can all be wrong. Responsible architecture makes uncertainty visible, permits abstention and stops unsupported work before it becomes action.
Human control and accountable delegation
Approval is meaningful only when a person sees the basis, understands the scope, can refuse and remains the owner of the decision.
Memory and proof
What a system records becomes part of future decisions. Memory therefore needs provenance, ownership, revision history and a way to correct error.
Public-interest technology
Some ideas should become operational systems. Other questions — about human agency, institutions and scrutiny — should remain open to public examination.
Publishing standard
Each published text identifies its author, version, publication and review dates, sources, AI assistance, editorial review and corrections path.