diff --git a/internal/db/migrations/0017_scan_runs.down.sql b/internal/db/migrations/0017_scan_runs.down.sql
new file mode 100644
index 00000000..a498146a
--- /dev/null
+++ b/internal/db/migrations/0017_scan_runs.down.sql
@@ -0,0 +1,2 @@
+DROP INDEX IF EXISTS scan_runs_started_at_idx;
+DROP TABLE IF EXISTS scan_runs;
diff --git a/internal/db/migrations/0017_scan_runs.up.sql b/internal/db/migrations/0017_scan_runs.up.sql
new file mode 100644
index 00000000..e17f4048
--- /dev/null
+++ b/internal/db/migrations/0017_scan_runs.up.sql
@@ -0,0 +1,20 @@
+-- M7 #381: track library-scan / MBID-backfill / cover-enrich runs so the
+-- admin overview can show last-run results + manual trigger state.
+CREATE TABLE scan_runs (
+    id              uuid PRIMARY KEY DEFAULT gen_random_uuid(),
+    started_at      timestamptz NOT NULL DEFAULT now(),
+    finished_at     timestamptz,
+    -- Per-stage tallies are jsonb so the schema doesn't need to grow when
+    -- a stage gains/loses a counter. Each stage's value matches the Go
+    -- struct emitted by that worker (e.g. library.Stats).
+    library         jsonb,
+    mbid_backfill   jsonb,
+    cover_enrich    jsonb,
+    -- Non-empty when the run as a whole errored (a stage panic, or the
+    -- worker hit a fatal error before writing stage tallies).
+    error_message   text
+);
+
+-- The status endpoint always reads the latest row; the in-flight guard
+-- needs to find rows where finished_at IS NULL. Both hit started_at DESC.
+CREATE INDEX scan_runs_started_at_idx ON scan_runs (started_at DESC);