The corpus asks back.
Jacaranda answers your questions. Lily constructs the ones your coverage hasn't answered yet — and hands them to your editors as assignments, evidence attached.
Questions are built, not imagined.
Lily reads the claims, the editorial positions, and the data series — and looks for tension between them: a capacity figure climbing while output stays flat; a promise dated last quarter with no delivery on record; your own hedged language clustering around one company's numbers.
Where the strata disagree, Lily constructs the question that disagreement implies. And a Lily question carries footnotes — every premise inside it traces to a record, so the question itself is auditable before anyone spends a day answering it.
Self-interrogation, on a schedule.
Lily asks. Jacaranda answers from the record — and states plainly where the record ends. That boundary is the product: “the archive contains no figure for X” is not a failure message, it's a commissioning brief with the news value pre-validated. The editor reads it, recognizes the thing only they know, and writes it. The story ingests. The corpus grows. Lily's next questions get sharper.
Decades of coverage, finally able to notice what it hasn't covered.
The system argues with itself — and logs the result.
Every Lily question doubles as a test. Jacaranda's answer is scored against the record by an automated audit; when an answer corrects a premise the question was built on, that mismatch is flagged — a stale claim found, a supersession missed. Quality stops being an impression and becomes a time series.
Jacaranda answers the questions you have. Lily asks the ones you don't.