diff --git a/image_analysis/scale_merge/SearchSpaceGroup.h b/image_analysis/scale_merge/SearchSpaceGroup.h
index da5519e7..4cb90d10 100644
--- a/image_analysis/scale_merge/SearchSpaceGroup.h
+++ b/image_analysis/scale_merge/SearchSpaceGroup.h
@@ -185,7 +185,21 @@ struct SearchSpaceGroupOptions {
     // real merohedral twins 1.82 and 4.01. 1.25 sat INSIDE the genuine range - four genuine promotions
     // already exceeded it and survived only because the two-arm rule happened to cover them, and one
     // cubic case with no such cover was refused outright by a margin of 0.4%.
-    double max_operator_h_ratio = 1.70;
+    //
+    // Re-measured over ~780 reports (three batteries and four de-novo integration-radius arms), H is the
+    // only Stage-A guard that ever fires: 14 refusals on three crystals, no chi-squared-ratio refusal and
+    // no b veto anywhere. Two of the three are the genuine trigonal twins, reading 1.90/1.91 on every arm
+    // and 3.61-4.80; the third is a weak tetragonal crystal reading 1.73-1.80 whose 422 is confirmed by
+    // every operator correlation and which three independent tests - our own L test on the merge it was
+    // demoted into, xtriage on the same merge ("most likely due to an NCS axis parallel to the twin
+    // axis"), and the arithmetic that a twin at the implied fraction caps pair correlation at 0.55 while
+    // the added operators score 0.84-0.90 - agree is not twinned. Its excess is the coverage limit named
+    // above: the added 422 operators exchange h and k and carry a systematic floor the parent's
+    // sign-flipping 2-folds do not. So the populations collide at 1.80 and 1.90 rather than at 1.70, and
+    // the bound sits at the midpoint. Degrading the data drives the ratio towards 1, never up, and the
+    // estimator is precise (bootstrap standard error 0.2-3.7% of the ratio), so the collision is a real
+    // overlap and not sampling noise - which is also why scaling the bound by that error cannot work.
+    double max_operator_h_ratio = 1.85;
 
     // The H test needs at least this many pairs on both sides to mean anything.
     int min_pairs_for_h = 200;