=== DF-4534 bounded hammer (guest boot #1, IRQ-driven lpt0, unpatched) ===
$ timeout 300 /root/close_lifetime 300
HAMMER-EXIT=0
child write results (typical, one per cycle):
[c] write -> 1024 errno=0        <- interrupted after 1 chunk drained, 1 chunk pending
[*] cycle 0 / 50 / 100 / 150 / 200 / 250
[*] done 300 cycles
dmesg after: no panic, no new messages beyond the per-cycle
"interrupt uses mplock" setup/teardown pairs
vmstat -m after:  lpt  Count 0, Requests 1.23K   (clean)
host-side port capture during hammer: 315 x 1024 'W' chunks continued
draining AFTER their write() calls had already returned (the callout
and IRQ chain finish interrupted transfers post-syscall) - these are
the same strobes whose in-flight edge is the UAF window.
Result: 300/300 cycles clean -> not_reproduced (silent-UAF class;
see VERDICT.md window quantification).