Nothing about your plant's care advice is improvised. Here's exactly how advice is put together before it reaches you.
Every plant in PlantAdvocate is matched to its species in the catalog: light needs, how dry the soil should get between waterings, humidity, soil, temperature, size, and toxicity. Each value in the researched catalog is cited to the extension service or botanical garden that published it, not written by a model. Older catalog values that aren't cited yet are labelled that way in the app, and the check schedules, which PlantAdvocate wrote for 129 common plants, are not cited.
Every watering, every repot, every check that ended in "still damp" builds a real history. This is the context no generic plant app has, because it's specific to your plant, its growing area, your habits.
This is the layer that does the reasoning. Its job is narrow on purpose: read the species record and the plant's history, then work out what to check and when — never to supply facts that aren't already in the record.
The Gnome tells you what your plant needs in plain English — specific to that plant, and built from its species record and its own log. Every species page shows which source each cited care fact came from.
An answer with no species record and no plant history behind it is a guess, however fluent it sounds. PlantAdvocate won't answer a question the data can't support.
When the catalog has no value for something, PlantAdvocate says it has no record instead of improvising.
The same species with different watering history gets different, specific guidance — not a one-size answer.
Every recommendation points back to what produced it, whether a catalog fact, your plant's care log or the local forecast. Each cited fact names its source, and anything not yet cited says so.