Resilience: Verdicts and the Results File
Each attack targets a resource, and that choice decides how a failure shows up. The memory attacks give the cleanest verdict (OOM-kill → exit 137) and chunk-size-overflow aims for a crash; the CPU attacks don't OOM, so they surface as hung or degraded through the liveness probe instead.
The verdict
For each attack the runner records a verdict from the server-side signal, plus peak RSS (vs. the idle baseline), the share of liveness probes that failed during the attack, and how long the server took to serve again after it:
| Verdict | Meaning | Pass |
|---|---|---|
survived | kept answering throughout, no OOM, memory stayed bounded | ✓ |
degraded | stopped answering under load but recovered afterwards | ✓ |
hung | still running but stopped answering and never recovered | ✗ |
crashed | the process died without being OOM-killed (panic/segfault) | ✗ |
oom-killed | the container ran out of its memory box | ✗ |
A liveness probe — a cheap GET on port 8080 — runs every second during the attack; recovery is measured by the same probe after the attack client stops. OOM is read straight from the container (State.OOMKilled / exit 137), and peak RSS from the cgroup's memory.peak.
The results file
A run writes results/resilience/<entry>.json (and, with --save, site/data/resilience/<entry>.json), keyed by the entry directory rather than the display name:
{
"framework": "aspnet-minimal",
"dir": "aspnet-minimal",
"passed": 4,
"total": 4,
"attacks": {
"slow-body": {
"verdict": "survived",
"passed": true,
"resource": "memory",
"peak_mem_mb": 71.2,
"baseline_mem_mb": 68.9,
"mem_limit_mb": 4096,
"liveness_fail_pct": 0.0,
"recovery_s": 0.0,
"oom_killed": false,
"client": { "established": 1000, "reaped": 1000, "open_at_end": 0 }
}
}
}
passed is the number of attacks the entry survived or only degraded under. The verdicts are not yet shown on the board or folded into the composite score — the suite records them for now, the way the TLS hardening checks did before they earned a badge.