diff --git a/CHANGELOG.md b/CHANGELOG.md
index 33c66d8d..e2ec54e0 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -19,6 +19,12 @@ Bug fixes:
     mismatched spots.
   - `generate_sim_spatialcluster`: check that the simulated dataset still lines
     up with the real one, through a new `check_alignment()` helper.
+  - `downstream`: reuse the `spatial_cluster` that
+    `process_datasets/generate_sim_spatialcluster` already computed, instead of
+    running `BayesSpace::spatialCluster()` a second time. Saves an MCMC run per
+    method per dataset and makes the metric report the clustering the pipeline
+    stored. This does not make `clustering_ari` reproducible: both before and
+    after, a fixed reference is compared against one unseeded run.
   - `generate_cosine()`: filter the Moran's I values rather than the objects
     holding them, so that a single NaN no longer takes the whole metric with
     it. `crosscor_cosine` was NA for 32 of 99 runs.
diff --git a/src/metrics/downstream/script.R b/src/metrics/downstream/script.R
index f03009c8..0f0dcb6e 100644
--- a/src/metrics/downstream/script.R
+++ b/src/metrics/downstream/script.R
@@ -43,13 +43,20 @@ crosscor_cosine <- generate_cosine(real_moransI, sim_moransI)
 crosscor_mantel <- generate_mantel(real_moransI, sim_moransI)
 
 cat("spatial clustering evaluation\n")
-# reclassify the clustering result
-real_cluster <- input_real_sp$obs[, c("spatial_cluster")]
-sim_cluster <- generate_sim_spatialCluster(input_real_sp, input_simulated_sp)
-location <- rownames(input_simulated_sp)
-sim_new_cluster <- reclassify_simsce(location, real_cluster, sim_cluster)
+# the simulated clustering was already computed by
+# process_datasets/generate_sim_spatialcluster; recomputing it here would give a
+# different answer every time, since BayesSpace is stochastic and unseeded
+real_cluster <- input_real_sp$obs[["spatial_cluster"]]
+sim_cluster <- input_simulated_sp$obs[["spatial_cluster"]]
 
-# ART and NMI
+if (is.null(sim_cluster)) {
+  stop(
+    "The simulated dataset has no 'spatial_cluster' column. ",
+    "Did it pass through process_datasets/generate_sim_spatialcluster?"
+  )
+}
+
+# ARI and NMI
 clustering_ari <- aricode::ARI(real_cluster, sim_cluster)
 clustering_nmi <- aricode::NMI(real_cluster, sim_cluster)