The interview (1.4.0)

The interview fills a model's gaps by asking the people who run the equipment, in short sessions on the Artekis/Interview page (/artekis/interview/<collection>, sized for a tablet) or in the context-model editor's Interview panel. The person names their role, answers up to five questions, and an admin saves what is worth keeping.

What it asks. Artekis picks the questions itself, from the coverage gaps an answer can close, in the editor's to-do order (open findings, then alarm activations in the last 7 days, then the lowest score). The writer model only puts them into words for the role. Documents, history and the operating state are never asked about.

Role	Asked about	One question per
Engineer	the engineering unit of a measurement tag with none	tag
Engineer	the normal range and alarm limits of a measurement tag with neither (an alarm setpoint counts)	tag
Engineer	which other equipment the asset feeds, controls, measures or shares a utility with	asset with no relation
Operator	what happens on the line when the asset goes wrong, and what they do	asset with no rule
Operator	what they check first when a measurement tag leaves its usual range	tag of an asset with no rule

A question that role skipped or answered in the last 30 days is not asked again, nor one whose answer is still waiting for an admin; when nothing is left the session does not start and says so.

Answers become drafts, never changes. A blank answer is a skip. When the person sends the answers, the writer drafts each one as an operating rule (condition, consequence, action; the answer itself always kept as the rule's text) or as a unit, a normal range or alarm limits on the question's tag; an answer that says nothing usable gives no draft. Every answer is kept in the session, drafted or not. Nothing reaches the model until someone at adminSecurityLevel saves: an admin who answered saves in the same step, anyone else's drafts wait under Waiting for an admin. Saving writes the ticked drafts, as edited, through the same checks as contextModel.put; a saved rule carries source: interview:<session id>, a unit or range lands on its tag, and provenance.lastInterview becomes today's date. Close without saving keeps nothing. After saving from the editor's panel, press Reload before editing the model above it.

Cost and limits. A session is one run on the Cost and Egress pages (template context-model-interview@1) with two writer calls: one phrases the questions, one drafts the answers. It is refused before its first call when its worst case is over $0.10 (with Claude Sonnet 5.5 as the writer it is about $0.08; a session measured $0.01), when the writer model has no price, under the collection's aiCutoff, or over the daily caps. Answers are at most 2,000 characters each and about 16,000 together; longer answers are refused before any call, so nothing is spent. Only the person who started a session, or an admin, answers it, and only once. On a tablet, the keyboard's own microphone dictates; Artekis has no speech code.