The AI-driven process behind the Stockyard models — methodology, research investigations, and what we learned along the way.
How AI discovers, tests, and improves the pitch grading models — from podcast transcripts to production upgrades.
About half of all tested features don't survive. We document them alongside the successes because the dead ends are often more instructive — CSW looked better than xRV by every per-pitch metric we measured, until we measured the right thing. Cascade+ was our most complex system ever, and direct regression beat it with a fraction of the infrastructure. The willingness to scrap significant investment when the data says so is what makes the process trustworthy.
