The empty-checklist failure
When the person doing the work picks the checks.
Organizations we have looked at closely had a version of this. If the person doing the work can choose which checks apply, acceptance becomes self-assessment with extra steps.
They pick the obligations they can meet, pass them, and get to say “all checks passed.”
The roster is fixed to the decision, not to whoever is doing the work. A shortened roster produces a refusal, not a smaller win.
All checks passed must never mean the caller selected the checks.
What the decision requires≠What the evidence achieves
What a decision requires is declared policy. What the evidence achieves is measured fact. An obligation is discharged only where achieved meets required on both strength of proof and independence, one requirement at a time.
Properties
Five things the roster must be.
Anchored
Bound to one decision and one defined set of things it applies to, so completeness is measured against a list that cannot quietly shrink.
Versioned
A decision made under roster v3 can still be read and understood after v5 ships. The roster has a lifecycle of its own, and its governance is half its value.
Non-empty
A roster with nothing on it is not a light-touch roster. It is an absent one.
Never shortened by the person doing the work
Obligations enter and leave through governance, never through the caller.
Separately declared
By an authority separate from the work. The party doing the work does not get to say what the work had to prove.
In regulated domains
Where written policy becomes executable.
The policy names the obligation, the roster carries it, the evidence classes discharge it, the verdict enforces it: a chain auditable end to end.
Many organizations find the relevant obligations are already distributed across policies, standards, contracts and runbooks. Whether those sources form a complete and maintainable roster for a named decision is a field question. The engineering opportunity is to compute against an anchored roster at reliance time rather than reconstruct it afterward.
A bounded diagnostic exercise
Take one decision your organization makes routinely and would struggle to defend afterward: a release to production, a figure in a filed statement, an approval, a batch release. Ask three things:
- Can the obligations for that decision be written down?
- Can the evidence classes that discharge them be named?
- Would that list stay stable long enough to be worth maintaining?
Where all three answers are provisionally yes, the decision is a candidate for governed finality; whether it works in practice still depends on the list being complete, the evidence being available, the authority being clear, conflicts being resolvable, and the list being maintainable.
Next steps
The roster declares. The evidence discharges.
What must be proved is domain knowledge and belongs with the people who hold it. What the standard fixes is the shape.
Typed evidence
What may discharge an obligation, what each class may claim, and how independence is measured.
Explore →NextAuthority and standing
Who may declare a roster, and why no surface grades its own work.
Explore →CoreThe Finality Kernel
How the roster and the evidence resolve into a verdict.
Explore →