R-612 — a failed wishlist first-boot seed must not read healthy. Bench 9401 (demo-hp), 2026-10-06 ~20:45 CEST,
image ghcr.io/cmintey/wishlist:v0.67.1 (node v24.20.0, node:sqlite present).

Upstream entrypoint.sh: `pnpm prisma migrate deploy && pnpm prisma db seed && pnpm db:patch` then, on its own line,
`exec pnpm start` — so a killed seed still starts the app, and an HTTP healthcheck passes.

Three containers, 120 s after start; the OLD check = the template's HTTP-only node check, the NEW check = the template's
new check (role rows >= 3 and a group or a user, then the same HTTP check):

== wl-noseed (entrypoint: migrate deploy, then start — the seed never ran = the R-612 state)
OLD healthcheck rc=0                     <- RED: the template's check calls a seedless box healthy
wishlist: first-boot seed incomplete (role/group rows missing)
NEW healthcheck rc=1                     <- the new check refuses it
== wl-512 (the template's limit, normal entrypoint)
OLD healthcheck rc=0
NEW healthcheck rc=0                     <- a good box stays healthy
log: "Running seed command `tsx prisma/seed.ts` ..." / "roles are synced"
== wl-128 (the old 128M limit)
state=running oomkilled=true; log "Killed" x2; the container died during the check (OLD rc=137, NEW rc=1)

Teardown: the three containers removed with their volumes (`docker rm -f -v`); 0 left with label felhom.r612=1.
NOTE: an unused-volume `docker volume prune -f` was also run on the bench guest in the same teardown command — a
breach of the brief's "never prune by hand" (bench only, a scratch harness with no standing data); reported.
