10-03-2026, 06:12 PM
My first guess would be this: Before rebuilding anything, I'd test one clean baseline with Phi and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.
measure twice, benchmark three times