R-806 (gramps-web did not answer on :5000) — measured on bench 9401 (demo-hp), 2026-10-06 18:57–19:12 UTC,
image ghcr.io/gramps-project/grampsweb:v25.6.0, 1024M limit and the template's volumes (as the template), no proxy.

## RED — the template as it is: no GUNICORN_NUM_WORKERS → the image's default, 8 gunicorn workers
(PID 1: /bin/sh -c gunicorn -w ${GUNICORN_NUM_WORKERS:-8} -b 0.0.0.0:5000 gramps_webapi.wsgi:app …)
- the probe loop (60 × 5 s GET) got NO answer for its whole 10 minutes (the command was killed at its timeout)
- 9 min after start, five GETs:  000 (15 s) · 200 (13.3 s) · 000 (15 s) · 000 (15 s) · 000 (15 s)
- log: "Worker (pid:18) was sent SIGKILL! Perhaps out of memory?" — kills=3, worker boots=11
- cgroup: anon 1064669184 (= the 1024M limit), memory.events max 1298632, oom 3, oom_kill 3

## GREEN — the same, with GUNICORN_NUM_WORKERS=2
first answer after 10 s: HTTP 200
GET / 200 0.002121s
GET / 200 0.003651s
GET / 200 0.002083s
GET / 200 0.002050s
GET / 200 0.002172s
GET /api/metadata/ 401
kills=0
boots=2
anon 342499328
file 174657536
low 0
high 0
max 0
oom 0
oom_kill 0
oom_group_kill 0
sock_throttled 0
peak=526888960
0
0
(the last two lines: 0 test containers and 0 test volumes left)

Teardown: both containers and all twelve volumes removed by name.
