diff --git a/upstream-submissions/kwin-fourier/kde-mr-body.md b/upstream-submissions/kwin-fourier/kde-mr-body.md
index 69348022fa..3d3e22373b 100644
--- a/upstream-submissions/kwin-fourier/kde-mr-body.md
+++ b/upstream-submissions/kwin-fourier/kde-mr-body.md
@@ -128,6 +128,25 @@ see the call site fire more often.
   Fence Poll Support" and match what `EXPORT_SYNC_FILE`+poll
   observes.
 
+### Subjective field-use note
+
+I have been running a more aggressive earlier version of this fix
+(simply *skipping* the `watchDmaBuf` wait — i.e. presenting without
+waiting for the producer fence at all) downstream as part of the
+fourier campaign on this hardware. Plasma feels measurably snappier
+under that variant; latency spikes during heavy compositor activity
+are gone. That earlier variant has different observable semantics
+than the patch in this MR (it can present unfinished frames if the
+producer is racing), so the subjective improvement *cannot* be
+directly attributed to the present patch — but the same hardware
+running the present patch does not regress against either the
+skip-the-wait variant or stock, so on this hardware/stack it's at
+worst a performance no-op and at best preserves whatever benefit
+the wait-skip variant was providing on the watchDmaBuf-firing path.
+
+Reviewers should treat this as anecdotal and weigh the patch on its
+correctness/cleanup merits.
+
 ## Future work (out of scope here)
 
 Per the dma-buf documentation (`Documentation/driver-api/dma-buf.rst`,