re: pre-register the widened F5 artwork test before reading the capture

F5's artwork half rests on 5 frames. This run presses A at the sweep's gate
instead of 40 frames later, giving ~40 frames of still-animating artwork,
and states the prediction for both outcomes before the log is opened.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Jc4pciRArGHfxGGhEbwp5t

docs/re/f5-artwork-window-prereg.md

@@ -0,0 +1,30 @@
+# Pre-registration — widening F5's 5-frame artwork window
+
+Committed **before** the capture is read. F5 concluded Ⓐ forces the plate and
+leaves the artwork's timeline alone, but the artwork half rested on **5 frames**
+(446–450 in `f5`): the only stretch where the press had landed and something was
+still animating. Everything else was at α255 and could not discriminate.
+
+**This run presses Ⓐ at the sweep's gate instead** — `PRESS2_FRAMES=3` rather
+than 40 — so the press lands at title `t≈70` with ~40 frames of artwork
+animation still ahead of it, an 8× wider window.
+
+## The prediction
+
+Comparing `f5-early` against the no-press control `f6b`, **aligned by the sweep's
+position**, over the four quads that are still ramping:
+`8ECA:-1.00`, `8ECA:-0.90`, `EAC3:-0.82`, `8154:-0.74`.
+
+| if Ⓐ … | those four |
+|---|---|
+| **only forces the plate** (F5's claim) | continue **identically** to the control for the full ~40 frames |
+| advances a shared clock | terminate early, jump, or change step |
+
+**A pass is the boring outcome and I have said so first.** The failure mode this
+guards against is that F5's artwork claim was read off a window too short to show
+a divergence that a longer one would reveal.
+
+## What it still will not cover
+
+The plate's identity, the 1.7× clock conflict, and whether Ⓐ drives some third
+clock — a cut is consistent with several mechanisms.