diff --git a/scripts/colab_status.py b/scripts/colab_status.py
index 608628f..2ba9a8c 100644
--- a/scripts/colab_status.py
+++ b/scripts/colab_status.py
@@ -70,6 +70,24 @@ def census(text):
                 sv = sorted(vals)
                 print(f"      {name:22s} n={len(sv):3d} median {sv[len(sv) // 2]:+7.2f} "
                       f"min {sv[0]:+7.2f} max {sv[-1]:+7.2f} violated {sum(1 for x in sv if x <= 0)}")
+    # A run of consecutive near-ties is a different finding from scattered ones: isolated lows are a
+    # hard tail, a contiguous block is a family of pairs the objective is not separating.
+    for epoch in sorted(per):
+        wins = sorted(w for w, v in by.items() if epoch in v)
+        best = run = 0
+        best_at = run_at = None
+        for w in wins:
+            if by[w][epoch] < 5.0:
+                if run == 0:
+                    run_at = w
+                run += 1
+                if run > best:
+                    best, best_at = run, run_at
+            else:
+                run = 0
+        if best >= 3:
+            print(f"      longest run of margin<5: {best} consecutive "
+                  f"(w{best_at}-w{best_at + best - 1})")
     both = [(w, v[1], v[2]) for w, v in sorted(by.items()) if 1 in v and 2 in v]
     if not both:
         print("  no window seen in both epochs yet")