its sounds like, tinfoil hat, they reduced the quant size of their model and tried to mask the change with the promo. your theory only addresses the spend not the reduced realiability