Ten LLM pods each rank ~1,500 US stocks every day from their own thesis; the top 100 are bought at the close and sold at the next open. Twenty random pods and five naive pods go through the same pipeline as the null. Primary metric: sector-neutral overnight excess return of the 100 picks versus the equal-weighted universe, in bps per day. No backtests: every day is a live forward test from 2026-09-22. A pod is called skilled only when its t-stat is at least 3 and it sits 2.33 sd above the random pods. Data as of 2026-09-22; 1 evaluated day.