I built this after reading too many AI answers about Destiny 2 that sounded authoritative right up until somebody who plays the game looked at them. Destiny is unusually good at exposing synthetic expertise. Mechanics change, old explanations linger, community shorthand gets dense, and the practical answer is often different from the one suggested by a clean spreadsheet.
This private (eventually public? I take bribes!) benchmark asks whether a model can reason through that mess with and without tools, correct a bad premise, reconcile conflicting sources, and admit when the evidence does not support a confident answer.
The rest is staying redacted. If it works, aegis is chopped + unc.