Sylpheed RE agent b69cc2325a re: quantify why the OB kill test keeps failing -- attackers are damaged, not missed ...

Fourth Stage 02 run: 520s of pilot with SYLPH_PREFER=e010 (198 fire=1), per-class
counts every 10s, OB+TIME throughout.

ARRIVAL TIMING n=4: 008 first at 02:05.00 (step bracket contains 120), 012 first
at 03:45.47 (bracket contains 210), then 012 held 3.5 more minutes with no fall
and no 016.

THE BLOCKER IS MEASURED, and it is not aim.  Reading hull at pos+0x154 for every
live e010 at the end: 14 of 16 are damaged, hulls 360..500, lowest 360/500 = 28%
gone.  The shots land; they just do not finish.  A kill needs roughly 3-4x longer
on one target than ~8 minutes of combat produced.

The contrast with the control shows the mechanism is hit points: e007 turrets
(HP 100) lost ~9 dead, 106 -> 97 live, while e010 attackers (HP 500) lost none and
the floor never moved off 16.

Decrement therefore still unproven after four runs, and "fly longer" is not a fix
since ~half of runs end early.  Options recorded for a redesign: point the pilot
at the WEAKEST attacker (hull is readable per entity; one is at 360/500), use
missiles (never deliberately fired), or wait on a scripted kill.

2026-08-27 03:07:33 +00:00

540 KiB

364x1248px

Raw History