Hold it up at a field and it tells you what is in it.
Crop, variety, drilling date, growth stage, and — if it is yours — the patch reference. It runs in the browser, it needs no app, and every camera frame is discarded on the device the moment it has been looked at.
What actually happens when you point it.
The phone works out where it is
Satellite positioning, to about five metres on a clear day. Worse under trees, considerably worse in the barn.
And which way it is facing
The compass and the accelerometer. This is the part that breaks near the grain store, which is full of steel.
It matches that against the field boundaries
Nine parcels, drawn from the land parcel identifiers the farm already files. Not invented for this.
It looks up what is in that field today
From the same crop record that drives the tracker and the seasonal state at the top of every page.
It draws the label
Crop, variety, drilled date, current growth stage, and whether any of it is yours.
And then it forgets the picture
The camera frame is never uploaded, never stored and never leaves the device.
What it can and cannot tell you.
| Point it at | It says | Reliability |
|---|---|---|
| One of the nine cropped fields | Crop, variety, drilled date, growth stage | Good, in the open, in daylight |
| A patch you hold | All of the above plus your patch reference and its bar | Good |
| The wildflower ground | What was sown, and this year’s pollinator count | Good |
| A hedgerow | Species recorded on that stretch, and when it was last laid | Reasonable |
| Livestock | Nothing. It does not identify animals and will not | Not attempted |
| Anything inside a building | Nothing useful. Steel and no satellites | Refuses rather than guesses |
| A neighbouring farm | Nothing. It stops at the boundary | Deliberate |
Why the camera frames never leave the phone.
An augmented-reality feature on a farm is a camera pointed at other people’s property, other people’s children and, occasionally, other people’s livestock. Uploading those frames to work out what is in them would be technically straightforward and completely indefensible.
So nothing is uploaded. The positioning happens on the device, the field lookup is a small request containing a coordinate and nothing else, and the camera image is drawn to the screen and discarded. There is no frame buffer, no cache and no diagnostic upload.
The cost of that decision is that it cannot recognise anything visually — it cannot tell you which weed you are looking at, and people ask for that constantly. Reading a field is a course rather than a camera feature, and it is a better answer anyway.
Five metres
On a clear day, in the open. Under the woodland it is closer to twenty.
Daylight only
It will run at dusk. It will not be useful.
Nothing uploaded
No frames, no video, no diagnostics containing images.
Nine fields
And it stops dead at the boundary of the farm.
It is a working farm and you cannot walk into most of it.
This feature works from the public footpaths, from the yard and from the car park, which is where you are allowed to be. It is not a licence to stand in the middle of a field of standing wheat in August holding a telephone. The footpaths are marked, the machinery is real, and the virtual walk exists precisely so that nobody has to do the dangerous version.
Pointing it, questioned.
No. It is a web page that asks for the camera and the compass, and if you refuse either it degrades to a map with the same information on it.
It works anywhere and says nothing anywhere else. We hold field boundaries for one farm.
A large steel building interferes with the compass. The label goes away rather than pointing at the wrong field, which is the correct failure.
No, and it is not going to. Identifying blackgrass from a phone photograph is a genuinely hard problem and getting it wrong costs a farmer real money.
The other half of this.
Point it at a can or a jar instead, and get the batch it came from.
