Temporal Probability Calibration
Published research on calibrating probabilistic predictions over time - cited by Google Brain, Amazon AWS, Stanford, NYU, and other leading institutions.
Written during the Sportsflare period.
Temporal Probability Calibration was written during our Sportsflare period, where probability quality mattered because predictions had to stay useful as events unfolded over time.
The paper studies calibration in temporal prediction settings. A model can be accurate in aggregate and still give probabilities that drift, overstate confidence, or become hard to trust at different points in a sequence.
That question shaped a lot of our later work: useful AI systems need more than a good answer. They need probabilities, uncertainty, and behaviour that remain reliable under time pressure.