diff --git a/apps/web/src/components/Nodes/frame/FrameNode.tsx b/apps/web/src/components/Nodes/frame/FrameNode.tsx
index bf06a13b4..8b707964f 100644
--- a/apps/web/src/components/Nodes/frame/FrameNode.tsx
+++ b/apps/web/src/components/Nodes/frame/FrameNode.tsx
@@ -510,7 +510,9 @@ export const FrameNode = memo(
           tooltipOffset={0}
           size={1}
           className={clsx(
-            'nodrag col-start-1 row-start-1 w-full min-w-0! bg-transparent px-1.5 text-xs font-medium outline-none',
+            // `text-ellipsis` only paints on an unfocused input, so the
+            // idle label shows "…" while editing still scrolls normally.
+            'nodrag col-start-1 row-start-1 w-full min-w-0! bg-transparent px-1.5 text-xs font-medium text-ellipsis outline-none',
             isEditingLabel
               ? 'text-fg-default cursor-text'
               : 'text-fg-muted hover:text-fg-default cursor-pointer',
diff --git a/docs/architecture/canvas-zoom-rendering.md b/docs/architecture/canvas-zoom-rendering.md
index 0c1038a92..c7e37e013 100644
--- a/docs/architecture/canvas-zoom-rendering.md
+++ b/docs/architecture/canvas-zoom-rendering.md
@@ -102,7 +102,7 @@ When frame labels collide, the higher-priority label wins:
 
 The first three interaction states force the affected label to remain visible and use matching overlay layers in descending order. With no interaction, the outer frame wins because zoomed-out views prioritize structural context over nested detail; the inner label returns after sufficient screen-space separation.
 
-Frame label width is capped to the transformed frame width with a 48 px usability floor. Overflowing names truncate visually while the input title retains access to the complete name.
+Frame label width is capped to the transformed frame width with a 48 px usability floor. An overflowing name truncates with an ellipsis so the clipped remainder is visible as such; the ellipsis disappears while the label is being edited, where the input scrolls instead.
 
 `FrameNode` owns the hierarchy and collision policy because it is frame-specific. `NodeWrapper` remains generic: it converts node coordinates to screen coordinates, applies owner-provided semantic visibility and width, handles interaction reveal, and performs opacity/FLIP transitions.