Exact NMOS 6502 cycle table replaces per-IR estimates

io_tick costs came from flat per-IR-op guesses (loads 4, reg-ops 2, max-per-address): INC zp ticked 4 vs the real 5, PHA 4 vs 3, (ind),Y loads 4 vs 5 — thousands of small errors per frame. On a machine where 76 cycles = one scanline and 1 cycle = 3 beam pixels, that error budget IS the human-reported symptom cluster: RESP-strobed sprites land pixels away from where game logic believes them (sprite drift + phantom CXPPMM collisions = random deaths), and…

Raw traces, ROM material, local paths, and personal data are not part of this public record.