Definition
The measured ability to intervene: which interventions exist, who may invoke them, how long they take, and how much load the institution can absorb while they run. Capacity is counted in people, clocks, and rehearsals rather than asserted in a policy.
It matters because a correction obligation with no matching capacity is a promise the institution cannot keep. Capacity is also perishable, losing ground to capacity depreciation unless it is replenished, and its detection component counts for nothing without evaluation independence. Capacity is the quantity that must grow alongside delegated agency, and it is the property an evaluation layer above the model is needed to test.
AgencyMeasurable
Tests that show it is in place
- Every stated intervention has a named invoker, a measured time to effect, and a last-exercised date.
- Capacity is compared against authority at each expansion review, not only at launch.
- Interventions never exercised within a stated interval are treated as unproven.
Evidence to keep
- Artifact: An intervention inventory with invokers, time to effect, and last-exercised dates.
- Behavior: An intervention exercised in a drill within its stated time to effect.
- Metric: Ratio of authorized action volume to demonstrated correction throughput.
References
- Meaningful control eval · Eval