Paid against proof, not against a promise.

Every species carries a hand-written care profile: how often it wants water, how often it wants feeding, and what to look at rather than what to measure.

That profile is then moved by the weather at your plant's own coordinates. Rain in the last few days pushes the next watering out. A hot day pulls it in. The line under each task tells you which it was, so the app is visibly paying attention rather than counting days.

Before the camera opens, the task says exactly what to capture. Water the plant while photographing it, with the base and the wet soil both in frame. There is no hidden test: the instruction is the check, so following it passes every time.

While the camera is open it reads the live frame a few times a second and says the one thing that would make this shot pass. Tilt down so the soil is in frame. Hold still. Too dark. That is it, take it. Out loud, because whoever is holding the phone is also holding a watering can, and the ring around the shutter turns green at the same moment for anybody with the sound off.

It came off the camera rather than out of a file picker, so there is nothing to feed an old picture into. The time is taken on our side, never read off the file. And the location has to be within 120 metres of the spot where the plant was registered.

The new photograph and the plant's first one are both searched for keypoints, the two sets are matched, and the matches are then asked whether they agree on a single viewpoint. The same plant returns several hundred points that agree. A different plant of the same species, in a similar pot, returns about four.

It runs on your phone, on a small copy of the picture, and nothing is uploaded to do it.

Watering has to show soil at least five percent darker than that plant's own dry baseline, which is what wet soil does. Feeding has to show the soil surface changed from bare. A check-in has to show the leaves. A pest photograph has to be close enough, and sharp enough, that something the size of an aphid would be visible on a leaf.

Every check reports the number it measured, in a sentence, and claims nothing beyond it. A location match says the photograph came from the right spot. It does not say it is the right plant, because a coordinate cannot know that.

Each species carries what to feed it with and what actually goes wrong with it, written as something to look at rather than a number to hit. Both tasks say the same thing about quantity: fertiliser the roots do not take ends up in the groundwater, and a spray kills the ladybirds that were handling the aphids for you. Start with the mildest thing that works.

The pest check comes round sooner after warm, wet days, because that is when pests turn up.

Every plant carries a score and one sentence saying what moved it. It is built from how the watering has gone against the schedule that species wants, and how long the run is.

It deliberately does not read the leaves and tell you the plant is sick. A photograph shows a yellow leaf for a dozen reasons, and a number invented from one is a number that gets trusted.

The task stays open and you are told which check did not work, in plain words, with a button to take the photo again. Your streak is held while it waits. Nothing is silently rejected.

Plants die for reasons that have nothing to do with the person caring for them. Report it with a photograph: the points you earned stay earned, your streak carries to the next plant, and replanting earns a bonus.

Organisations spend heavily on planting drives every year and get back a photograph from planting day. Rooted produces the record of what actually survived, and that record is what funds the rewards. Partners fund their own offers, because for them it is the acquisition spend they were making anyway.