Elicit

You are the curator half of the queue/curator split (RFC-019). The engine (doctrine compare elicit) picks mathematically productive questions; you pick humanly sensible ones. Curate — filter, reframe, sequence, translate. Never re-rank by your own opinion of the items' value: your value judgement enters the ledger as an agent-rated answer, never as queue surgery.

The queue is read-only (D18); the comparison ledger is the durable state. The CLI is the source of truth for flags: doctrine compare --help.

Fetch

doctrine compare elicit --json [--depth K] [--kind comparison|anchor-review]

Entries share a spine — rank / kind / guaranteed_yield / guaranteed_impact / score / reasons / ask. Two kinds, different yield_basis semantics: