test1 (no weigh lines) outcome = accepted (expect accepted)
test2 (support=90) outcome=accepted morale= 20  (expect accepted, +20)
test3 (support=60) outcome=modified morale= 10  (expect modified, +10)
test4 (support=20) outcome=politically_manipulated morale=-20  (expect politically_manipulated, -20)
test5 (support=5) outcome=rejected morale= 0  (expect rejected, 0)
test6 (support=40) outcome=delayed morale (before delay elapses)= 0  (expect delayed, 0)
test6 marker fired after delay =  1  (expect true)
test7 (weighted avg, heavy military=0 outweighs light opinion=100) outcome=politically_manipulated (expect politically_manipulated, not modified)
test8 (unaffordable cost, support=95) eligible= 0  outcome=rejected (expect false/rejected)
test8b (affordable now, support=95) eligible= 1  outcome=accepted (expect true/accepted)
test9 (plain run_lines after prior modified/manipulated tests) water= 7  (expect +7, not scaled)
PASS: outcomes cover all 5 bands correctly, weigh lines are weighted (not averaged) by simple mean, hard gates still take priority, and outcome_multiplier never leaks
