Different expertise needs different grain; grain should be switchable, not one size

What it is

A newcomer wants a briefing “a bit shorter.” An expert wants the subject of the third sentence in that same briefing made passive, and two terms locked. What the two people need is not a better default grain. It is grain that can change gear inside one product. Control granularity should be switchable: coarse is a few intent buttons, fine is sentence, word, parameter, and the person changes gear mid-task rather than having it locked to an account type.

One size offends both sides. Everyone coarse, and the expert has nothing to point at. Everyone fine, and the newcomer has not learned what those objects are. The failure is not picking the wrong gear. It is forbidding a change of gear.

Why it happens

Expertise is not a fixed trait of a person. It is relative to the current object. The same person is a newcomer on mail and an expert on their own lab notes. Assigning grain from a “user persona” mis-assigns it when the task switches. Grain also tracks the artefact: before there is a draft, coarse intents help; once the draft has structure, a coarse intent will smash that structure.

An unswitchable UI makes grain a mode, and a wrong mode wastes the whole operation. Switchable grain makes it a current tool; a wrong gear can be changed back at once. The cost of trial sits on the gear, not on the whole artefact.

Studying it

Have the same people do two tasks: rhetorical revision outside their domain, precise revision inside it. Interface conditions: locked coarse / locked fine / switchable mid-task. Dependent variables: success on each task, how often and when switches happen (start, after failure, after the artefact has formed), time spent in the wrong gear.

Do not group by self-report “am I an expert” — that measures identity. Use task performance or a domain test. The critical look is whether the switchable group walks to fine on the in-domain task and stays coarse on the out-of-domain one. If they do not, the switch entry itself is undiscoverable.

Where it stops holding

A single-task, single-audience tool (fixed fill-ins on an internal ticket) does not need gears; one gear is honest. If a switch has no state cue, people will not know which gear they are in, and fine-grain parameters will leak into a coarse session. A mobile screen cannot hold every fine object; the switch can be “open desktop-grade editing” rather than stacking another layer on the same glass. An audio-only channel cannot show fine objects; the object has to be confirmed before grain is changed.

Applying it

• Offer coarse intents (shorter, different reader) and fine objects (sentence, word, field) beside the artefact. Start coarse by default; once someone clicks into a sentence, open fine without a trip to Settings to change mode.
• The current gear must be visible and reversible in one action. Do not replace this-task grain with an account-level Advanced mode.
• Switching to fine must not clear coarse constraints; they should stack: first “for patients,” then change the subject of sentence three.
• Check: have the same person first compress out of domain, then revise sentence by sentence in domain. Locking either gear makes one of those clearly worse; under a switchable condition both should complete, and logs should show at least one person-initiated gear change. If none happens, move the entry to the row under the caret, not the far end of the toolbar.

Related

• Same group: L2.13.1 If the output cannot be edited, the user can only accept it or start over · L2.13.2 After the user edits the result, later turns must respect those changes · L2.13.3 Too many control dimensions turn a generation tool into specialist software that has to be learned
• Nearby: L2.06 Granularity of Control · L2.04 Complementarity of Parameter Controls and Natural Language · L2.03 Examples and Template Guidance
• Search terms: switchable control granularity · expertise reversal · mode versus tool grain