diff --git a/docs/re/METHOD.md b/docs/re/METHOD.md
index 8f634072..c5ddc973 100644
--- a/docs/re/METHOD.md
+++ b/docs/re/METHOD.md
@@ -1600,6 +1600,14 @@ one bounding box.
 This produced two wrong findings in one session, one of them reported to another
 agent with three alternative explanations "ruled out":
 
+⚠️ **It then produced a THIRD, 2026-09-02, with this entry already written.**
+`f6-unit5` / `f6-unit6` declared `pteff03a` absent from every capture when it was
+the second quad of an `indices=8` draw the whole time. Two sections of this file
+and a `REFUTED.md` line already said a draw carries more than one quad. The trap
+is not that the fact is unknown -- it is that **a reader written before consulting
+this file reproduces the bug the file exists to prevent.** Use
+`tools/re-capture/read_draws.py` rather than rolling a fresh regex.
+
 * **`ptlogo_back2eff3` "is never drawn by the game".** It is batched with
   `ptlogo_back2eff4`, and because the wipe family is right-aligned, `eff3`
   (788…1196) lies **entirely inside** `eff4` (447…1196). The union is *exactly*
diff --git a/docs/re/f6-two-leaf-periods-confirmed.md b/docs/re/f6-two-leaf-periods-confirmed.md
index 550e57b1..0edcfb32 100644
--- a/docs/re/f6-two-leaf-periods-confirmed.md
+++ b/docs/re/f6-two-leaf-periods-confirmed.md
@@ -47,12 +47,20 @@ was obtained here without a new capture: **both leaves give a complete
 boundary-to-boundary cycle in both existing logs.** The old reader saw one quad,
 so it could see only one of the four cycles on disk.
 
-## Refutation attempt
+## Refutation attempt — made, and it **did not land**
 
-Against the port: they now draw both sweeps, and **if they run both leaves at one
-rate the strips will stay locked and drift apart from the game within a single
-loop** — ~118 units of separation per cycle, growing. Flagged; I cannot test
-their renderer from here, so this is a prediction for them to check, not a result.
+Against the port: they now draw both sweeps, so if they ran both leaves at one
+rate the strips would stay locked and drift ~118 units per cycle.
+
+**They already run two timelines.** Measured on their side — the right instrument
+for a question *about* their renderer, per R5: at a raw leaf clock of 4873 the two
+leaves read **72.6 and 552.6** against **73 and 553** from their own declared
+spans of 600 and 720, over 17 748 probe samples. The span is taken per leaf from
+that leaf's own keyframes, so they were never locked.
+
+Recorded as an attempt that failed. Still worth making: they report they would
+otherwise have asserted it from reading the code, and the property had never been
+measured.
 
 ## Not settled