Topic overview · Verifiable automation
The boundary is not where a system can act, but where its outcome stays checkable.
Responsible automation carries its own verifiability. It separates production, control and responsibility — and stops where judgement is no longer possible.
Three task classes, three boundaries
Judging automation by feasibility is not enough. The measure is: can we determine, at reasonable cost, that an outcome is correct, complete and fit for purpose? The less verifiable an output, the smaller its permitted effect should be.
1
Deterministically testablePatterns, required fields, totals, tests — machine-decidable.
autonomous · scoped
2
Assessable, not computableRubrics, reference examples, sampling, traceable approval.
prepare + review
3
Judgement-dependentEthical, strategic, legal consequences — not formalisable.
human decides
01 · Criteria firstA verification contract before the run: outcome, sources, observable acceptance, prohibited deviation.
02 · Verification layersRule, independent review, source check and human share the control.
03 · Review loadProduce less, group deviations, prioritise risk — so judgement stays possible.
Core method
The verification card
Task & effectWhat is produced, and what may change as a result?
Task classDeterministic, assessable or judgement-dependent?
Observable criteriaHow would two people recognise the same acceptable outcome?
Verification layersWhich rule, test, source and person will check it?
Review loadHow many outcomes can be inspected attentively?
Stop & approvalWhen does automation end, which decision remains human?
Break a recurring process down so production, verification and responsibility become visibly distinct — and set two stop signals before you start.