The read-back check cannot tell two identical commands in a row apart:
the done bits are latched and the stale word differs in no bit from the
new one. Alternating bit 29 per command would have closed that blind
spot, but only if the engine executes commands with the bit at either
polarity, and every word ever observed from firmware has it set.
This probe answers the question with read-only SPD traffic: a baseline
run with the bit set, the same reads with it clear, then an alternating
run. Measured 2026-07-28: exit 3, the engine consumes GO but sets the
error bit for every command with bit 29 clear, 8/8, while interleaved
reads with the bit set keep working. The bit is must-be-set, not a
toggle, so the alternation scheme is dead and the blind spot is
documented at imc_command_landed() instead. The probe stays as the
evidence behind that paragraph.
