diff --git a/cmd/csv_real_purchase_test.go b/cmd/csv_real_purchase_test.go
index a3deb8aae..90def6f85 100644
--- a/cmd/csv_real_purchase_test.go
+++ b/cmd/csv_real_purchase_test.go
@@ -103,6 +103,8 @@ type csvPurchaseEnv struct {
 	// expectPurchaseFailure makes run require errPurchasesFailed (#2157):
 	// a real run with any failed purchase now returns it so the CLI exits non-zero.
 	expectPurchaseFailure bool
+	// cfgMod adjusts the Config run builds, e.g. to mimic the flags of a real invocation.
+	cfgMod func(*Config)
 }
 
 // newCSVPurchaseEnv wires the stub with the RDS and ElastiCache offering and
@@ -135,6 +137,9 @@ func newCSVPurchaseEnvFor(t *testing.T, csv string, fixtureSets ...map[string]aw
 		recs, err := loadRecommendationsFromCSV(csvPath)
 		require.NoError(t, err)
 		cfg := Config{CallerAccountID: defaultTestAccount, CSVOutput: filepath.Join(dir, report), AuditLog: filepath.Join(dir, "audit.jsonl")}
+		if env.cfgMod != nil {
+			env.cfgMod(&cfg)
+		}
 		err = runCSVPurchases(context.Background(), awsCfg, recs, false, cfg, "run-"+report)
 		if env.expectPurchaseFailure {
 			require.ErrorIs(t, err, errPurchasesFailed)
@@ -392,3 +397,27 @@ func TestLoadRecommendationsFromCSV_MemoryDBEngineIsCanonical(t *testing.T) {
 		assert.Equal(t, &common.CacheDetails{Engine: "redis", NodeType: "db.t4g.small"}, recs[0].Details, "engine=%q", engine)
 	}
 }
+
+// TestCSVRealPurchase_IDLabelsAppliedCoverage pins #2248: the generated id's
+// sizing label must match the coverage the CSV run applied. Without --coverage
+// the flag default (80) is unused and the rows are bought at 100%. RDS uses the
+// generated id as ReservedDBInstanceId; the ElastiCache client names its own.
+func TestCSVRealPurchase_IDLabelsAppliedCoverage(t *testing.T) {
+	tests := []struct {
+		name string
+		cfg  func(*Config)
+		want string
+	}{
+		{"no --coverage buys 100%", func(c *Config) { c.Coverage = 80; c.CoverageFromDefault = true }, "-100pct-"},
+		{"explicit --coverage 100", func(c *Config) { c.Coverage = 100 }, "-100pct-"},
+		{"target-coverage wins", func(c *Config) { c.Coverage = 80; c.CoverageFromDefault = true; c.TargetCoverage = 100 }, "-100pct-"},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			env := newCSVPurchaseEnv(t)
+			env.cfgMod = tt.cfg
+			env.run("report1.csv")
+			assert.Contains(t, env.purchaseForm(t, rdsHost+" "+rdsPurchase).Get("ReservedDBInstanceId"), tt.want)
+		})
+	}
+}
diff --git a/cmd/multi_service.go b/cmd/multi_service.go
index 3f7cb312d..0fb0eb9e1 100644
--- a/cmd/multi_service.go
+++ b/cmd/multi_service.go
@@ -772,6 +772,10 @@ func runToolFromCSV(ctx context.Context, cfg Config) error {
 // them, and writes the report and summary for whatever ran. A non-nil error
 // is errAuditTrailLost and/or errPurchasesFailed.
 func runCSVPurchases(ctx context.Context, awsCfg aws.Config, recs []common.Recommendation, isDryRun bool, cfg Config, runID string) error {
+	// Purchase IDs label the sizing via cfg.Coverage (effectiveSizingPct);
+	// pin it to the coverage the CSV sizing used so the label cannot drift (#2248).
+	cfg.Coverage = determineCSVCoverage(cfg)
+
 	// Create account alias cache for lookup
 	accountCache := NewAccountAliasCache(awsCfg)