Most research is frozen the day it's published. Ours runs in a loop: the robots train every day, the records change, the gates re-fire, and the papers evolve — with every revision graded against eight objective gates before it ships. Every number on every page traces to a real engine record you can open. When our results move, the paper says so before we do.
Every paper on this shelf is a living artifact. New training results land in the records → the loop wakes → every claim is re-verified → the weakest section (and only that section) is rewritten → eight gates grade the revision → the version number climbs. Nothing here is ever "finished." It's currently winning.