The Strange Data Project.

Most AI demonstrations run on data the models have already seen a great deal of. The interesting question is what happens at the edges — with the records a sector keeps and nobody else does, the formats nobody standardised, the material that never made it into a training set.

That is where the real work sits for most businesses, and it is the part a generic benchmark tells you nothing about.

We documented what worked, what didn't, and what surprised us, and rolled the lessons into how we train clients. An open competition is a cheap way to find out something true, and a much better use of a prize fund than a marketing campaign of the same value.