diff --git a/.changeset/benchmark-run-count.md b/.changeset/benchmark-run-count.md
new file mode 100644
index 0000000..9fe31e1
--- /dev/null
+++ b/.changeset/benchmark-run-count.md
@@ -0,0 +1,5 @@
+---
+"tech-stack": patch
+---
+
+Benchmark comparisons need at least 5 interleaved runs per side, up from 3.
diff --git a/skills/tech-stack/references/practices/testing.md b/skills/tech-stack/references/practices/testing.md
index f4e2a0f..a4e3a4a 100644
--- a/skills/tech-stack/references/practices/testing.md
+++ b/skills/tech-stack/references/practices/testing.md
@@ -35,4 +35,4 @@
 - Record an engine-exit diagnostic that cannot be fixed in an exact-match allowlist: message, resource path, count. Every other ERROR/WARNING fails.
 
 ## Performance comparisons
-- Interleave baseline and candidate runs on the same machine, at least 3 runs each (Alárinà requires 5). Report median and p95 with ranges; serial batches can turn time-dependent noise into a false regression.
+- Interleave baseline and candidate runs on the same machine, at least 5 runs per side. Report median and p95 with ranges; serial batches can turn time-dependent noise into a false regression.