The idea
Join sports modelling, reporting and audio publishing into one coherent editorial system.
The connection
Forecasting and storytelling solve different problems. The model must be evaluated against outcomes; the editorial layer must explain the result without turning statistical uncertainty into certainty.
System design
The following structure makes the project concept inspectable and defines the implementation to demonstrate.
What would prove it
Keep a timestamped record of forecasts made before games, then compare outcomes using calibration and a declared scoring rule. Separately check whether written and spoken summaries preserve the underlying numbers.
The next complete milestone
A defensible next milestone is a replayable reporting day: source snapshot, model version, pre-game forecast, published copy and post-game evaluation. Attractive commentary alone does not establish predictive performance.
Project record
An autonomous NBA analytics and editorial operation running on a Cloudflare Worker: calibrated Elo models, a daily written brief, a generated podcast, and a magazine-style front page, all updating without a human in the loop.
Source: the published APX project register, recovered 28 September 2026. This study explains the design direction and evaluation path; it is not a fresh audit of the product implementation.