diff --git a/docs/re/f5-snap-target-undecodable-with-reach.md b/docs/re/f5-snap-target-undecodable-with-reach.md
new file mode 100644
index 00000000..ef67b7f1
--- /dev/null
+++ b/docs/re/f5-snap-target-undecodable-with-reach.md
@@ -0,0 +1,60 @@
+# F5 — the snap target is `[160,238)` and **cannot be pinned from any capture**
+
+**Question:** what title time does Ⓐ jump the clock to?
+
+**What the human looks at:** nothing — this is a negative, and its value is that
+it stops anyone spending another run on it.
+
+**What this does NOT cover:** the per-cycle phase fit, F6.
+
+**Classification: ❔ undecodable, with reach.** ⟨capture⟩ + ⟨disc⟩.
+
+## The bound tightened to `[160,238)`
+
+`ptcopyright.t32` declares `0(α0) · 138(α0) · 160(α255) · 238(α255) · 244(α0)`,
+and in the press run it goes **absent at f435 → α255 at f436**, skipping its
+whole 22-unit ramp. So **t ≥ 160**.
+
+Its identification is measured, not assumed: in the no-input control it first
+draws at **t≈143** and reaches α255 at **t≈164**, against a declared 138→160.
+
+The sweeps are at α255 at the same frame, and their parents' exit runs 238→250,
+so **t < 238**.
+
+## Why it stops there — nothing on the screen changes inside that window
+
+Every element in build 4 with a keyframe past `t=100`, and where it is static:
+
+| element | keys | static across |
+|---|---|---|
+| `ptlogo1` / `ptlogo2` | 0,26,34,38,40,42,**251**,264 | 42→251 |
+| `ptlogo_tm` | 0,98,**116**,**242**,248 | 116→242 |
+| `pteff00.prm` | 0,**16**,**261**,269 | 16→261 |
+| `ptloop01` / `ptloop02` | 0,70,**100**,**238**,250 | 100→238 |
+| `pteff02.prm` | 0,46,61,76,**118**,**236** | 118→236 (α0 both ends) |
+| `ptlogo_back2` | 0,66,**80**,**243**,249 | 80→243 |
+| `ptlogo_back2eff` | 0,62,**66**,**238**,244 | 66→238 |
+| `ptcopyright` | 0,138,**160**,**238**,244 | 160→238 |
+| `pteff01`, `ptlogo_back2eff5`, `ptlogoall_eff`, `ptlogoall_eff2` | last key ≤118 | — |
+
+**Their intersection is exactly `[160,238)`.** The game draws a bit-identical
+frame at `t=160` and at `t=236`. No capture can separate them, because there is
+nothing left that moves.
+
+## Refutation attempt — the port's `236.0`
+
+I tried to refute it and **cannot**, but the finding is stronger than "survives":
+**236.0 can never be confirmed or refuted by a capture.** Every value in
+`[160,238)` is observationally identical, so their choice is free.
+
+Nor does it become observable later: the clock **freezes** at settle
+([`ui-clock-freezes-at-settle.md`](structures/ui-clock-freezes-at-settle.md)), and
+the `238…250` exit plays **when the screen leaves, not on a timer** — so a
+different starting `t` inside the window produces no difference at any point.
+
+## Not settled
+
+* The exact value, permanently, from this route. Only guest memory or the code
+  that performs the assignment could give it, and that is a different instrument
+  than any capture.
+* The per-cycle phase fit for the post-wrap decomposition.