Why we built PlantPhrase.
Tomato growers already have the data. What they lack is something that reads it the way an agronomist would.
A grower with a sensor in the bed can tell you today's temperature and humidity. What that does not tell them is whether early blight pressure has been quietly accumulating for eleven days, or whether the vapour pressure deficit has been sitting above the flowering band long enough to matter. Those answers exist in the agronomic literature. They have mostly not existed in software you can point at your own plants.
Most plant apps do one of two things. They identify a species from a photo and stop there, or they draw a chart of temperature over time and leave the interpretation to you. Neither closes the loop between what is happening and what to do about it.
Published models, not vibes
Every number PlantPhrase shows traces to a published model. Tom-Cast severity values come from the Pitblado table for Alternaria solani, with the same temperature bounds the field research uses. Growing degree days use the capped method at base 10 °C. Vapour pressure deficit is Tetens, computed leaf-to-air rather than the simpler air-only shortcut that overstates how comfortable a plant is.
This matters because the alternative is a model that cannot be checked. If a system tells you to spray and cannot say which threshold it crossed, you have no way to know whether it was right, and no way to learn when it was wrong.
Honest about what it does not know
The harder discipline is refusing to answer. A vision result below the confidence threshold returns uncertain with retake guidance instead of a plausible-sounding guess. A confident result the system cannot map to a known disease is also uncertain, with the raw label recorded so the gap is measurable rather than invisible.
The same applies to missing data. Accumulating models need continuous readings, and a day without readings is skipped rather than filled with zeros. A zero-filled day looks healthy on a dashboard; a missing one looks missing. Short sensor gaps are bridged by interpolation, longer ones can draw growing degree days from historical weather, and leaf wetness is never inferred at all, because inventing wetness would mean inventing disease pressure.
Where it is now
PlantPhrase is in private beta, focused on tomatoes. Tomatoes are a deliberate constraint rather than a limitation we plan to apologise for: the disease models are crop-specific, the thresholds are field-validated for this crop, and a system that is genuinely correct about one plant is worth more than one that is vaguely plausible about a hundred.
If you are growing this season and want to put it in front of real plants, get in touch or request a spot.