Users need different explanation levels in different situations

What it is

The same screening model: a duty nurse has to decide whether this alert should wake a doctor; a department head has to decide whether to keep it next quarter. The first needs an account locked to this instance; the second needs an account locked to overall tendency. Users need different explanation levels in different situations: the right level follows the judgement now being made. There is no explanation that is always enough.

What is missing is not word count. It is the pairing of level with judgement.

Why it happens

Local and global compress different kinds of information and can support different acts. The instance level supports accepting, changing an input, or contesting this item. The overall level supports continued reliance, narrowing the use, or switching off. People switch between those judgements in the same interface, sometimes seconds apart — handle this alert, then ask at handover whether the thing was too noisy tonight. If the interface only ever supplies one level, the other judgement will be made with the wrong material.

The cue for the switch is the task, not the job title. The same person is an operator by day and a procurement reviewer by night. One manual issued by employee number will be at the wrong level in one of those situations.

Studying it

Build two task blocks: dispose of a single output, and decide whether to keep using the system in that setting. Everyone does both blocks, but receives only one level (local only or global only), or a level that switches with the task. Independent variables: explanation level, whether level switches with task, task order. Dependent variables: quality of the instance disposition, calibration of the continue-to-use decision, count of cross-level errors (arguing “the system is reliable overall” from a single heatmap, and so on).

The point is the cross-level error, not whether “having an explanation” feels more satisfying.

Where it stops holding

Some settings only ever have one judgement — a suggested reply in a support toolbar almost never becomes a procurement decision, so local alone is fine. Some settings bind both judgements to one control (“adopt and make this the default”), so both levels must be given and labelled apart. This entry does not prescribe how to write local or global copy; it only requires that level follow situation.

Applying it

• Pick the default level from the current act: local when handling this item; global when setting, granting permission, or turning the thing on.
• Allow an explicit switch: “about this item” / “about this system.” Do not melt the two levels into one paragraph.
• At handovers, weekly reviews, and off-switch confirmations — entries that make an overall judgement — do not only replay the latest local account.
• Check: intercept the moments of “dispose of this item” and “decide whether to keep using it,” and see whether the copy on screen can support that judgement. If a head of department is holding one heatmap while discussing renewal, the level has not followed the situation.

Related

• Same group: L5.02.1 A local explanation accounts for a single output · L5.02.2 A global explanation accounts for overall behavioural tendency · L5.02.4 Local explanations support accepting or contesting a single result · L5.02.5 Global explanations support judging whether to keep relying on the system · L5.02.6 An overall impression induced from a few local explanations is often wrong · L5.02.7 Feature importance is a correlation ranking, not a causal account · L5.02.8 Developers need debugging information; end users need grounds for action — they cannot share one explanation
• Nearby: L5.01 Types of Explainability · L5.05 The Moderation Principle of Transparency · L5.08 Counterfactual Explanations
• Search terms: explanation level · situation-dependent explanation · local versus global