Real-world risk
Every domain we touch is the same activity — reading signal out of noise to support decisions that carry consequences.
Super Typhoon Sinlaku · Earth observation
The decisions that matter most are made on noisy, incomplete evidence. We're an independent research collective doing the inference in the open — rigorous work, the data and code beneath it, and honest limits on what it shows — so the people making the call have something solid to stand on.
Knowledge narrows uncertainty. We do the narrowing in public.
We sit between a lab and a consultancy: research-grade in method, applied in intent — across safeguarding, insurance, natural catastrophe and AI. Every paper is built to be read by a practitioner facing a real decision, not only cited by another paper. Where we use AI to generate data or accelerate analysis, we say so, with provenance, validation and limitations on the table.
Every paper is built from one source into two artifacts — a citable PDF and a fast reading view — with the data and code beside it.
Safeguarding, insurance and risk, natural catastrophe, and artificial intelligence — four fields where consequential decisions are made under uncertainty. The same inference method reads signal from the data in each: from online-harm disclosure, to hazard and loss, to the behaviour of the models themselves.