Kill criteria
Pre-committed, quantified conditions under which an automated strategy or agent is halted without further discretionary judgement — specified before deployment and binding on the operator.
The position
Almost nothing is written on this, and it is the most load-bearing control in autonomous finance. The reason is uncomfortable: a real kill criterion removes discretion from the person who would otherwise exercise it, at exactly the moment they most want it. Criteria written after deployment are not criteria, they are rationalisations with a timestamp. The test of whether an institution has genuine kill criteria is whether anyone has ever been overruled by them.
This section is a claim, not a definition. It is argued rather than asserted, and it is the part of this page you are invited to disagree with. Corrections and counter-arguments to [email protected] are published.