Another great thing about Oaken: as I rip out old models, I have so much less finagling to do because with Oaken’s seeds, I only have a small dataset that’s logically broken up. So I nuke the old seeds, work on a single test until it passes; then the whole suite.