The standalone guest checks process-wide suspend/resume with three busy threads and one blocked sibling, blocked SIGCONT, repeated stop/continue notifications from wait4 and waitid, killing a stopped process, nice, graceful SIGTERM cleanup, and unprivileged signal/priority errors. Concurrent waiters verify that one reaps and the other receives ECHILD; a 40-child test checks that every child can wake a blocking wait. Every target is a child created by the test.

Build the desired kernel architecture first, then boot its regression:

python3 tests/activity-monitor/run.py --repo /Users/alex/code/vinix --kernel /tmp/worktree/kernel --arch aarch64
python3 tests/activity-monitor/run.py --repo /Users/alex/code/vinix --kernel /tmp/worktree/kernel --arch x86_64

--repo supplies the existing boot dependencies and userland sysroot; --kernel supplies the already built isolated kernel. The desktop's owned-child action tests are in desktop/tools/tests/activity_controls_test.v.

metrics.c checks advancing machine and per-core CPU counters, regular-file and positioned logical I/O accounting, and INET UDP loopback payload counters. /proc/stat uses USER_HZ=100. /proc/activity_io reports completed physical disk bytes and INET payload bytes, including loopback and excluding Unix sockets and protocol overhead. /proc/<pid>/io uses rchar/wchar for logical regular-file bytes, read_bytes/write_bytes for physical I/O attributed to the submitting thread, and net_recv_bytes/net_send_bytes for payload. Background writeback is not assigned to the original writer. Physical completion accounting covers AHCI, NVMe, ATA, VirtIO and Apple ANS storage.

The allocation allowlist records two Apple SMC device objects and two Resource registrations: the existing battery and the new battery-power device. They are created once and retained by their global pointers and devtmpfs for the kernel's lifetime; repeated reads do not create devices. The installed V compiler also reports allocation categories missing from the repository's older allowlist and its x86 warning-only source scan has pre-existing import/interface errors. A baseline comparison must distinguish those from new allocation sites rather than accepting all warnings. The complete ARM scan introduces only the owned battery-power object and registration; the new polling Text builders remain on the stack. The guest regression measures every monitored procfs path directly.