diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index aa2ef1b..40454c5 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -20,3 +20,16 @@ jobs:
           shellcheck -x "${files[@]}"
       - name: cli tests
         run: bash test/cli.sh
+
+  # Kept SEPARATE from `check` on purpose: this job pulls a Postgres image and
+  # stands up throwaway containers, and a slow image pull must never delay the
+  # fast shellcheck + cli.sh feedback above. ubuntu-latest ships Docker running
+  # and passwordless sudo, so test/db-integration.sh EXECUTES here (it only
+  # skips where Docker is absent). It is the automated proof that dump/restore
+  # actually round-trips, not just that the args parse.
+  db-integration:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: db dump/restore round-trip
+        run: bash test/db-integration.sh
diff --git a/README.md b/README.md
index f16497c..34da5ce 100644
--- a/README.md
+++ b/README.md
@@ -187,6 +187,44 @@ gunzip -c <artifact> | docker exec -i <container> sh -c \
   non-emptiness *before* the prompt and before anything touches the database, so
   a fat-fingered path fails cheaply.
 
+#### Verifying a dump/restore actually works
+
+A `.gz` that opens without error is not proof of a good backup: a `pg_dump`
+truncated mid-stream still compresses into a perfectly valid gzip file that
+*looks* exactly like a complete one. The same ethos as the nightly dump applies
+here — **a backup you have never read back is not yet a backup.** The only
+fully-trustworthy proof is to restore the artifact and read the rows back out.
+
+On a real Coolify box you can do that **without touching prod data** by
+restoring into a fresh *scratch* database rather than over the live one:
+
+```sh
+# 1. dump the live database (read-only; harmless)
+rig db dump coolify-db /srv/snapshots/verify.sql.gz
+
+# 2. create a throwaway database as the container's OWN superuser
+docker exec coolify-db sh -c 'createdb -U "$POSTGRES_USER" rig_verify'
+
+# 3. restore the artifact INTO the scratch db (not the live one)
+rig db restore /srv/snapshots/verify.sql.gz coolify-db rig_verify --yes
+
+# 4. spot-check a table you expect to see
+docker exec coolify-db sh -c \
+  'psql -U "$POSTGRES_USER" -d rig_verify -c "\dt"'
+docker exec coolify-db sh -c \
+  'psql -U "$POSTGRES_USER" -d rig_verify -c "SELECT count(*) FROM <some-table>"'
+
+# 5. drop the scratch db — live data was never touched
+docker exec coolify-db sh -c 'dropdb -U "$POSTGRES_USER" rig_verify'
+```
+
+If the counts and tables are there, the artifact is real. This is exactly the
+round-trip `test/db-integration.sh` automates in CI (the `db-integration` job):
+it seeds a known table in a source container whose superuser is *not* the
+default, dumps it, restores into a second container whose superuser differs, and
+asserts the rows and an ordered checksum survived — the same proof, done against
+throwaway containers on every push.
+
 ### `rig runner install --repo <owner/repo>`
 
 Runner box only, run after `rig bootstrap runner` (the same two-step rhythm
diff --git a/test/db-integration.sh b/test/db-integration.sh
new file mode 100755
index 0000000..996b218
--- /dev/null
+++ b/test/db-integration.sh
@@ -0,0 +1,178 @@
+#!/usr/bin/env bash
+# test/db-integration.sh — a REAL dump/restore round-trip for `rig db`.
+#
+# test/cli.sh proves the arg parsing; this proves the actual thing works. It
+# stands up throwaway PostgreSQL containers, seeds a known table, runs the real
+# `rig db dump` / `rig db restore`, and reads the rows back out the far side —
+# because a dump piped through gzip can look perfectly valid while being
+# truncated, and only restoring it and reading it back proves otherwise.
+#
+# It exercises the two invariants db.sh is built around:
+#   * the source superuser is a NON-default name (src_super), so a green run
+#     proves the code reads the CONTAINER's own $POSTGRES_USER/$POSTGRES_DB and
+#     never a hardcoded `postgres`;
+#   * the destination superuser is a DIFFERENT name (dst_super), so the restore
+#     only succeeds because --no-owner --no-acl stripped the source role graph —
+#     a plain dump would abort under ON_ERROR_STOP=1 on the first missing role.
+#
+# Skips cleanly (exit 0) when it cannot run — no Docker, no reachable daemon, or
+# no way to become root (rig db requires root) — so it never reddens a dev box
+# that simply has no Docker. On ubuntu-latest CI, Docker is preinstalled and
+# running and passwordless sudo works, so it EXECUTES for real there.
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+RIG="$ROOT/bin/rig"
+PG_IMAGE="${RIG_DBIT_PG_IMAGE:-postgres:16-alpine}"
+
+PASS=0 FAIL=0
+ok()   { echo "ok: $*";   PASS=$((PASS + 1)); }
+bad()  { echo "FAIL: $*"; FAIL=$((FAIL + 1)); }
+skip() { echo "skip: $*"; exit 0; }
+die()  { echo "FAIL: $*" >&2; exit 1; }   # trap still runs cleanup
+
+# --- privilege / docker preflight -------------------------------------------
+# rig db requires root (require_common_guards). CI's runner user is not root but
+# has passwordless sudo and docker-group access; mirror that exactly. EVERYTHING
+# that touches Docker or rig goes through as_root so containers and the artifact
+# share one owner and cleanup is uniform (root can always reach the socket).
+if [ "$(id -u)" -eq 0 ]; then
+  as_root() { "$@"; }
+elif sudo -n true >/dev/null 2>&1; then
+  as_root() { sudo "$@"; }
+else
+  skip "not root and no passwordless sudo — rig db requires root; cannot run the round-trip here"
+fi
+
+command -v docker >/dev/null 2>&1 || skip "docker not installed — nothing to exercise"
+as_root docker info >/dev/null 2>&1 || skip "docker daemon not reachable — skipping the live round-trip"
+# Pull up front so an offline box SKIPS (not FAILS): a missing network is not a
+# regression in rig. On CI the image is fetched here and the run below is fast.
+as_root docker pull "$PG_IMAGE" >/dev/null 2>&1 || skip "could not pull ${PG_IMAGE} (offline?) — skipping"
+
+# --- unique names + guaranteed cleanup --------------------------------------
+PREFIX="rig_dbit_$$"
+SRC="${PREFIX}_src"
+DST="${PREFIX}_dst"
+WORKDIR="$(mktemp -d)"
+ARTIFACT="$WORKDIR/roundtrip.sql.gz"
+
+cleanup() {
+  as_root docker rm -f "$SRC" "$DST" >/dev/null 2>&1 || true
+  as_root rm -rf "$WORKDIR" >/dev/null 2>&1 || true
+}
+trap cleanup EXIT
+
+# A stale run must never collide with this one.
+as_root docker rm -f "$SRC" "$DST" >/dev/null 2>&1 || true
+
+# --- helpers ----------------------------------------------------------------
+# Poll until <container> answers a real query as <user>/<db>. pg_isready alone
+# reports "ready" during the entrypoint's temp-server phase, so gate on an
+# actual SELECT succeeding instead. Bounded (~60s) so a wedged box fails, loudly.
+wait_ready() { # <container> <user> <db>
+  local c="$1" u="$2" d="$3" i=0
+  while [ "$i" -lt 60 ]; do
+    if as_root docker exec "$c" psql -U "$u" -d "$d" -tAc 'SELECT 1' >/dev/null 2>&1; then
+      return 0
+    fi
+    i=$((i + 1))
+    sleep 1
+  done
+  die "container ${c} never started accepting connections as ${u}/${d}"
+}
+
+# An ordered, checksummable fingerprint of the fixture table — the thing that
+# must survive the round-trip byte for byte.
+fingerprint() { # <container> <user> <db>
+  as_root docker exec "$1" psql -U "$2" -d "$3" -tAc \
+    "SELECT id||'|'||name||'|'||qty FROM widgets ORDER BY id" | md5sum | cut -d' ' -f1
+}
+rowcount() { # <container> <user> <db>
+  as_root docker exec "$1" psql -U "$2" -d "$3" -tAc 'SELECT count(*) FROM widgets' | tr -d '[:space:]'
+}
+
+# --- source: NON-default superuser, seeded fixture --------------------------
+as_root docker run -d --name "$SRC" \
+  -e POSTGRES_USER=src_super -e POSTGRES_DB=src_appdb -e POSTGRES_PASSWORD=srcpw \
+  "$PG_IMAGE" >/dev/null
+wait_ready "$SRC" src_super src_appdb
+
+as_root docker exec "$SRC" psql -U src_super -d src_appdb -v ON_ERROR_STOP=1 -c "
+  CREATE TABLE widgets (id int PRIMARY KEY, name text NOT NULL, qty int NOT NULL);
+  INSERT INTO widgets VALUES (1,'alpha',10),(2,'beta',20),(3,'gamma',30);
+" >/dev/null
+
+# assert_eq <desc> <want> <got>
+assert_eq() {
+  if [ "$2" = "$3" ]; then ok "$1"; else bad "$1 — wanted [$2] got [$3]"; fi
+}
+
+SRC_FP="$(fingerprint "$SRC" src_super src_appdb)"
+assert_eq "source seeded with 3 known rows" 3 "$(rowcount "$SRC" src_super src_appdb)"
+
+# --- dump (explicit outfile) ------------------------------------------------
+if as_root "$RIG" db dump "$SRC" "$ARTIFACT" >/dev/null 2>&1; then
+  ok "rig db dump exited 0"
+else
+  bad "rig db dump failed"
+fi
+if as_root test -s "$ARTIFACT"; then
+  ok "dump wrote a non-empty artifact"
+else
+  die "dump produced no usable artifact — nothing to restore"
+fi
+
+# --- dump (default outfile naming: <container>-<UTC-timestamp>.sql.gz) -------
+( cd "$WORKDIR" && as_root "$RIG" db dump "$SRC" >/dev/null 2>&1 )
+if compgen -G "$WORKDIR/${SRC}-*.sql.gz" >/dev/null; then
+  ok "dump with no outfile wrote <container>-<timestamp>.sql.gz in cwd"
+else
+  bad "default dump name not produced"
+fi
+
+# --- destination: DIFFERENT superuser (the portability proof) ---------------
+as_root docker run -d --name "$DST" \
+  -e POSTGRES_USER=dst_super -e POSTGRES_DB=dst_appdb -e POSTGRES_PASSWORD=dstpw \
+  "$PG_IMAGE" >/dev/null
+wait_ready "$DST" dst_super dst_appdb
+
+# --- restore into the default database --------------------------------------
+if as_root "$RIG" db restore "$ARTIFACT" "$DST" --yes >/dev/null 2>&1; then
+  ok "rig db restore exited 0 across a differing superuser"
+else
+  bad "rig db restore failed (a role-graph leak would abort here)"
+fi
+
+DST_FP="$(fingerprint "$DST" dst_super dst_appdb)"
+assert_eq "destination has exactly 3 rows after restore" 3 "$(rowcount "$DST" dst_super dst_appdb)"
+if [ "$DST_FP" = "$SRC_FP" ]; then
+  ok "restored fingerprint matches source ($SRC_FP)"
+else
+  bad "restored data differs — src=$SRC_FP dst=$DST_FP"
+fi
+
+# --- idempotency: --clean --if-exists means a second restore is a no-op-ish --
+if as_root "$RIG" db restore "$ARTIFACT" "$DST" --yes >/dev/null 2>&1; then
+  ok "second restore exited 0 (dump is --clean --if-exists)"
+else
+  bad "second restore failed"
+fi
+assert_eq "still exactly 3 rows after re-restore (no duplication)" 3 "$(rowcount "$DST" dst_super dst_appdb)"
+assert_eq "fingerprint stable after re-restore" "$SRC_FP" "$(fingerprint "$DST" dst_super dst_appdb)"
+
+# --- restore into a NAMED scratch db (the [db] arg / shared-Postgres path) ---
+# This is exactly the safe manual proof the README documents: restore into a
+# fresh scratch database rather than over live data.
+as_root docker exec "$DST" createdb -U dst_super rig_verify >/dev/null
+if as_root "$RIG" db restore "$ARTIFACT" "$DST" rig_verify --yes >/dev/null 2>&1; then
+  ok "rig db restore into a named [db] exited 0"
+else
+  bad "restore into named database failed"
+fi
+assert_eq "named-database restore reproduced the source fingerprint" \
+  "$SRC_FP" "$(fingerprint "$DST" dst_super rig_verify)"
+
+echo "---"
+echo "$PASS passed, $FAIL failed"
+[ "$FAIL" -eq 0 ]