Compare an intervention that can answer the question

Consider a read-only setup assistant. It must express a requested duration in seconds, obtain the current permitted range and report the converted value, rule revision and whether the request is within that range. The duration interpretation recurs; the permitted range can change. The runner supplies arithmetic operations and a current-rule read. Both remain available through the comparison.

Two guides, a and b, offer the same duration interpretation through different entries. The following are stipulated probabilities of completing the whole report correctly on the selected task population, with policy, routing, task distribution and other support fixed.

Whole-guidance versus no-guidance comparison gives a difference of 0.70. Removing only a gives zero; removing only b also gives zero. Each zero answers a question about the arrangement containing the other guide. Removing both produces 0.20. If the receiving floor is 0.85, retaining either guide is adequate under these assumptions, whereas joint removal is not. After removing one, re-establish the comparison before removing the other, or test the proposed joint composition directly.

Now consider an aid with 0.20 correct completion both present and absent. Its zero difference accompanies inadequate work. Examine whether it was retrieved, available before the decision, understood and usable; whether its instruction was correct; and whether another required contribution was missing. That comparison supplies no grounds for declaring the operation acquired. At a high adequate rate, the same zero could instead make removing that aid worthwhile, subject to the remaining joint conditions.

For an operating comparison, select the intervention before collecting its data. A whole-bank comparison measures that bundle. A one-guide removal needs the same policy, task weighting and remaining composition; replacing the missing retrieval with another guide changes the intervention. If several guides or the router change, compare the resulting arrangement. Preserve the cases on which a returned instruction was ignored or misleading. Use SYSE.49/.46 to obtain and qualify the outcome basis, including uncertainty that can change the decision. Adaptive observations can choose the next intervention; reserve a separate final comparison for the reliance placed on the selected arrangement.

The primary techniques make different choices at this point. SKILL0 v2, §3 ranks guidance under a budget scheduled to reach zero; its whole-bank and individual-removal formulations need to be distinguished when implementing the comparison. A zero endpoint can remove still-useful help. SLIM v2, §3 instead uses exposure, smoothed removal differences and repeated low contribution before retirement, and can add guidance for persistent failures. Freeze the comparison composition, or deliberately assess the changed bundle after each mutation. Neither rule makes individual retirement proof of acquired capability or protects it from later forgetting.