Re-running one of our best stock-picking workflows: using GPT-5.2 to spot information asymmetry in public markets

In our previous run of this workflow, OpenAI’s GPT-5.2 in Deep Research mode recommended $MP, and we observed strong early performance (approximately +30% in the 13 days following the recommendation). The purpose of this experiment is simple: if the workflow works, that success should be repeatable.

Accordingly, we re-ran the same methodology under the same constraints and required the model to select a single winner using explicit expected-value analysis.

This memo documents the process and results end-to-end.