diff --git a/.research/experiments/frontier-workload-regime-taxonomy-20260720.md b/.research/experiments/frontier-workload-regime-taxonomy-20260720.md
index 6bb6eeb..72360ee 100644
--- a/.research/experiments/frontier-workload-regime-taxonomy-20260720.md
+++ b/.research/experiments/frontier-workload-regime-taxonomy-20260720.md
@@ -90,7 +90,7 @@ I believe arrival bursts primarily affect waiting/admission and tail TTFT/E2E, w
 Discovery uses Qwen30B because its 12-cell `TP x MNS` surface already has simulator and real anchors:
 
 - TP: `{1, 2, 4}`
-- MNS: `{8, 32, 64, 128}`
+- MNS: `{8, 16, 32, 64}`
 - objectives: mean/p90 TTFT, TPOT, E2E
 
 Qwen235B is a held-out confirmation, not pooled into discovery: