test1 (no weigh lines) weighed =  0  support =  0  (expect 0/0)
test2 (one weigh line, public_opinion=73) weighed =  1  support =  73  (expect 1/73)
test3 (weighted 75/25 mix of 80 and 20) support =  65  (expect 65, NOT the plain mean 50)
test4 (10x repeated calls) opinion unchanged =  1  confidence unchanged =  1  eventlog unchanged =  1 
test5 preview support =  55  (expect 55), real outcome = modified (expect modified, since 50<=55<70)
PASS: preview_support_score matches determine_outcome's real formula exactly and never mutates anything, even across repeated speculative calls
