Case studies
Real pictures, put through the app as it is.
Four public sets of thermal pictures. For each one: what the app said, where it was right, where it was wrong, and what it could not tell.
Read this first
None of these plants is a PanelProof customer. Their owners and the people who took the pictures took no part in this. The pictures are public, and we ran them ourselves.
The four sets
Rooftop plant, Catanzaro
8 full photos, 1,485 panels. 162 evenly warm panels marked by comparing neighbours. No technician's marks to score them against.
Read itOne array, Louisiana
3,675 panel pictures. 406 of 409 pattern faults flagged. 375 of 431 evenly warm panels missed.
Read itThe test it could not study for
3,002 panel pictures. 92 of 100 faulty panels found. 39 of 1,000 normal ones flagged.
Read itSolar farm, Sindos
2,772 panels from 47 photos. No fault marks exist. 7.8 of every 100 panels sent to a person.
Read itHow to read them
What a case study is
A run, not a story
Each set went through the same code a customer's upload goes through: the file reader, the grid, the fault finder, the review rule and, on full photos, the neighbour check. Nothing was tuned for the set, and nothing was left out afterwards.
Where the people who made a set marked its faults, we show the app against those marks. Where nobody did, we say so, and we show only what the app said.