diff --git a/harness/run.go b/harness/run.go
index fee15f9..2d07c24 100644
--- a/harness/run.go
+++ b/harness/run.go
@@ -279,11 +279,22 @@ func (m *RunManager) runScenario(ctx context.Context, runID string, sc *Scenario
 	aborted := ctx.Err() != nil
 	// revert (always, even on abort, unless the scenario is irreversible by design or skip_revert)
 	if !skipRevert && sc.Revert.Method != "none" {
-		rvCtx, cancelRv := context.WithTimeout(bg, 10*time.Minute)
-		rres, rerr := a.runStep(rvCtx, sc, sc.Revert, runID, params)
-		cancelRv()
+		// a failed revert leaves the fault in place (2026-10-18: one SSH connection reset at the X-M2 revert left a
+		// netem qdisc on a legacy instance and broke the next 14 runs of two scenarios), so retry a few times
+		var rres map[string]any
+		var rerr error
+		for attempt := 1; attempt <= 3; attempt++ {
+			rvCtx, cancelRv := context.WithTimeout(bg, 10*time.Minute)
+			rres, rerr = a.runStep(rvCtx, sc, sc.Revert, runID, params)
+			cancelRv()
+			if rerr == nil {
+				break
+			}
+			a.store.Event(bg, &runID, &sid, "note", "runner", map[string]any{"phase": "revert", "attempt": attempt, "error": rerr.Error(), "result": rres})
+			time.Sleep(15 * time.Second)
+		}
 		if rerr != nil {
-			a.store.Event(bg, &runID, &sid, "note", "runner", map[string]any{"phase": "revert", "error": rerr.Error(), "result": rres})
+			a.store.Event(bg, &runID, &sid, "note", "runner", map[string]any{"phase": "revert", "error": rerr.Error(), "result": rres, "note": "revert failed after 3 attempts; the fault may still be in place"})
 		} else {
 			a.store.Event(bg, &runID, &sid, "note", "runner", map[string]any{"phase": "revert", "result": rres})
 		}