What we measure, what we predict, and what we do not claim. In plain English.
What goes in
Five sensor channels per node, the local microclimate, the variety planted, and the harvest outcome at season end.
- Soil moisture, NPK, pH, soil temperature, ambient humidity. Sampled every 15 minutes during the growing season.
- Microclimate priors from regional weather stations and SRTM elevation tiles.
- Harvest data is the slow loop. One row per (field, season, variety) at harvest.
- No personal data leaves the farm by default. Variety and outcome rows are opt-in for the data marketplace.
What comes out
Per-field, per-season ranked variety candidates with a predicted yield delta, a confidence band, and a vendor link. Plus the existing irrigation and fertiliser triggers.
- Predicted Δ yield: a point estimate with a low / medium / high confidence label.
- Vendor: a marketplace link to a seller that has the variety in stock for your region.
- Confidence drops sharply when fewer than three pilot farms in your microclimate have grown the variety.
How it is validated
We check our predictions against public variety trials, and against pilot fields we hold back each season.
- A field we test against is never used to make that season’s prediction.
- Public trial data corrects for breeders quoting their best-case numbers.
Source:
AHDB Recommended Lists ↗
What it does not do (yet)
The limits, up front.
- No causal yield guarantees. Δ yield is a statistical estimate, not a warranty.
- No insurance-grade certification. We do not yet meet the actuarial standards for a parametric crop-insurance trigger.
- No automatic field action without grower confirmation. Auto-irrigate triggers a partner integration; the grower keeps the override.
- No livestock telemetry, no grain marketing, no tractor ISOBUS. These are not in scope.
