7503066a91
With the value filter in place, an attach watching turret losses -- frequent, unlike marked-fighter kills -- narrowed from 374 candidates to 2 to 0 across three events. That is exactly how a correlation search should behave, and it ends in a refutation rather than fizzling out: no plain u32 in the entity heap decrements consistently with turret kills. Witness checked first, per the rule adopted last iteration: 12 of 25 samples stalled, but all four kill events fall in the early non-stalled stretch between t=45 and t=177, so the events themselves are sound. The negative fits the objective text rather than contradicting it. Stage 02 phase 1 asks to shoot down all invading enemy fighters, and turrets are not fighters, so a counter of remaining marked targets should not move when a turret dies. It also rules out the cheaper reading that REMAINING OB is a general kill or enemy-remaining tally: it ignored ten turret deaths. The method is proven and the search region is right, since 0xbdb59668 where the counter was originally seen lies inside the scanned range. What is missing is unchanged -- two or three e010 kill events in non-stalled samples -- against the same two obstacles, the freeze rate and a pilot managing about two marked-fighter kills per five minutes. One clean 220 s run this iteration produced zero.