diff --git a/docs/re/f6-two-leaf-periods-prereg.md b/docs/re/f6-two-leaf-periods-prereg.md
new file mode 100644
index 00000000..a9e6dadb
--- /dev/null
+++ b/docs/re/f6-two-leaf-periods-prereg.md
@@ -0,0 +1,20 @@
+# Pre-registration — the two sweeps must loop at 720:600
+
+Committed **before** measuring. `pteff03a` was only found to be drawn at all in
+[`f6-unit11`](f6-unit11-pteff03a-IS-drawn.md), so its declared timeline has never
+been tested against a capture.
+
+The leaves declare **different loop lengths**: `pteff03` **600 units**,
+`pteff03a` **720**. If that decode is right the two strips must drift apart, with
+
+> **period(pteff03a) / period(pteff03) = 720 / 600 = 1.200**
+
+measured inside a single capture, so no clock or frame rate enters.
+
+| if the decode is right | if it is wrong |
+|---|---|
+| ratio ≈ 1.20, in both captures | equal periods (one shared clock), or any other ratio |
+
+Equal periods is the outcome that would flatter a lazy reading — one traveller
+driving both quads — and it is exactly what unit 11's batched-draw discovery
+could plausibly have produced. So the test can fail against me.