diff --git a/docs/active-work.md b/docs/active-work.md
index 00bb8fe..7db8ac3 100644
--- a/docs/active-work.md
+++ b/docs/active-work.md
@@ -319,7 +319,7 @@ Framing
 this base, the panel-replay prerequisite recorded as DISCHARGED by #243,
 and the both-axis effect witness still owed.
 
-**Landed so far** (latest verified code head `ec6444e`)**:** B1's per-target fractional
+**Landed so far** (latest verified code head `269c52e`)**:** B1's per-target fractional
 wheel residual (the producer), B2's daemon-side horizontal panel leg,
 B3/B7's shared `scroll_window_columns` with its saturated bound and wrap
 pin, B4's middle-click PRIMARY paste, **B6's minimap wheel routing**,
@@ -457,21 +457,41 @@ blank, the text off its left edge. `clamp_code_scroll_left` now sits at
 `reshape`'s tail, beside B5's icon hook and for the same reason.
 
 **Owed to the framing, not to the code: L2's wording**, which promises a
-witness this frontend cannot provide. The GPU's four `manual_left_authority`
-writes are still read by nothing; whether they should be deleted or given
-a reader is a framing decision, not one to take in passing.
+witness this frontend cannot provide.
+
+**Settled (user decision, 2026-09-02): the GPU latch is deleted, not
+completed.** The contract is behavioral, and the two frontends are not
+required to share a representation; a reader would have duplicated the
+painted-before policy and needed a cursor baseline of its own to avoid
+suppressing genuine cursor movement. `manual_left_authority`, its
+initializer and its four writes are gone, and `scroll_by_columns`
+no longer returns an unused `bool`. **GPU authority is structural**,
+and the framing now says so.
+
+The GPU's four rows: **L2** (height-only preserves — the policy, not a
+latch), **L7a** (widening clamps to the exact bound), **L7b** (a
+content shrink clamps **through the incremental edit path**), and
+**L3** (a moved `CursorByte` pulls the viewport back). Each fires on
+its own mutation; L2's also necessarily bites L7a, which its doc names.
+
+**L7b, GPU, is the one that found a live gap.** Q#R1's keystroke case
+re-shapes only the affected line through `try_reshape_line` and skips
+the full `reshape` — **and skipped clause 3's clamp with it**. A
+one-line delete shortening the widest line could leave the viewport
+past the end of the text with no later event to repair it. The clamp
+now runs on that branch too.
 
 **Landed but NOT yet witnessed:**
 
-- **wrap and buffer-replacement clearing** of that latch and the origin
-  (clause 5), on both the GPU's `scroll_by_columns` and the buffer
-  replacement path;
 - **R4 and R5's residual resets**, as two separate clears beside
   `code_scroll_left` so omitting either is individually visible.
 
-**Owed outright:** L7b and the GPU's whole read side (above), step 3's
-fractional both-axis panel witness, R4/R5's own replacement witnesses,
-and **B1's disposal half** — a residual keyed to a surface that goes
+(The wrap and buffer-replacement clearing that stood here is done: the
+TUI's four replacement paths carry L8b–L8e, wrap carries L8, and the
+GPU half of that bullet described the latch that no longer exists.)
+
+**Owed outright:** step 3's fractional both-axis panel witness, R4/R5's
+own replacement witnesses, and **B1's disposal half** — a residual keyed to a surface that goes
 away must go with it, and this frontend does not yet track "that buffer
 is gone".
 
@@ -557,12 +577,14 @@ error: could not compile `pmacs-gpu` (bin "pmacs-gpu" test)
        due to 4 previous errors
 ```
 
-**THREE occurrences now**, all local. The third (at `9cb610e`) gave the
+**FOUR occurrences now**, all local. The third (at `9cb610e`) gave the
 complete set of four sites, which the second's captured tail had cut to
 three: at that head they were `main.rs:9049`, `:13546`, `:14772` and
 `:14833` — every use of the module the 1b branch added in `9e54cd2`.
-Line numbers drift as the file grows; the module path is the stable
-part of the signature.
+The fourth (2026-09-02, mid-edit) hit the same four uses at
+`:9109`, `:13658`, `:14884` and `:14945`. **Line numbers drift as the
+file grows and are not part of the signature; the module path is**, and
+the count of four is stable across all of them.
 
 An earlier draft of this entry guessed the missing fourth site was
 `widest_display_columns`. The third occurrence shows that guess was
diff --git a/docs/gui-stage1-input-framing.md b/docs/gui-stage1-input-framing.md
index bb7a7e0..7f46060 100644
--- a/docs/gui-stage1-input-framing.md
+++ b/docs/gui-stage1-input-framing.md
@@ -841,9 +841,39 @@ released snaps to the caret's.
 **L2 is height-only for the same reason L3 leaves the viewport.** A
 resize that changes width also changes the clamp, so the origin could
 move for a reason unrelated to authority and the row would not
-discriminate. Height-only invokes the real follow path
-(`resize` → `ensure_caret_painted` → `horizontal_follow`) while leaving
-horizontal geometry fixed.
+discriminate.
+
+**CORRECTION (revision 21, implementation).** This paragraph said
+height-only "invokes the real follow path
+(`resize` → `ensure_caret_painted` → `horizontal_follow`)". **That is
+false in the setup L2 itself requires.** `resize` runs
+`ensure_caret_painted` only when the caret was painted, and L2's caret
+is deliberately outside the manual viewport — so Q#F6's painted-before
+policy skips the follow, and no follow runs at all.
+
+**The consequence is that the GPU needs no authority flag.** The TUI
+does: `horizontal_follow` runs on every paint there and would drag the
+origin back, so only a latch can outrank it. On the GPU the follow
+reaches the origin only through `ensure_caret_painted`, which is
+skipped exactly when the user has scrolled the caret off screen — and
+when the caret IS on screen it is inside the viewport, where
+`follow_left` returns the origin it was handed. **GPU authority is
+structural: there is no follow to outrank.**
+
+A flag was carried there for a while, written in four places and read
+in none. **It has been removed rather than completed** (user decision,
+2026-09-02): giving it a reader would have duplicated the
+painted-before policy and needed a cursor baseline of its own to avoid
+suppressing genuine cursor movement. **The contract is behavioral; the
+two frontends are not required to share a representation.**
+
+So the GPU's rows are: **L2** (height-only preserves, witnessing the
+policy — mutation: follow unconditionally), **L7a** (widening clamps to
+the exact bound), **L7b** (a content shrink clamps **through the
+incremental edit path**, which bypasses `reshape` and so bypassed the
+clamp), and **L3** (a moved `CursorByte` pulls the viewport back —
+the "until the cursor changes" boundary, which without a flag is the
+only thing that marks it).
 
 **L7 needs both legs.** Clause 3 promises re-clamping on *geometry and
 content* changes; widening the viewport witnesses only the geometry