diff --git a/README.md b/README.md
index d4575cf..dd78866 100644
--- a/README.md
+++ b/README.md
@@ -90,15 +90,20 @@ resizing the panel, and the surround is teal (`-solid`) rather than the root's b
 **The stage is small because the panel scales it by whole numbers.** Pixel art at a fractional scale gets
 uneven cells — some three screen pixels wide, their neighbours two — which reads as "not quite right"
 rather than as blur, and `image-rendering: pixelated` does not help because unevenness is not blur. So the
-panel snaps the scale, in DEVICE pixels, to whole numbers below 2x and QUARTER steps above it. Quarters are
-safe because the unit that has to stay regular is the CELL: the board is a 16px grid, so at 2.25x a cell is
-exactly 36 screen pixels, as even as at 2x, and only sub-cell pixels alternate. They are quarters rather
-than halves because coarse steps overshoot — with halves this jumped 2x to 2.5x, too small to too large
-with nothing in between.
+panel snaps the scale, in DEVICE pixels, to QUARTER steps. Quarters are safe because the unit that has to
+stay regular is the CELL: the board is a 16px grid, so every quarter step gives a whole cell — 20px at
+1.25x, 36px at 2.25x — and only sub-cell pixels alternate. Quarters rather than halves because coarse steps
+overshoot, and quarters all the way down rather than whole numbers under 2x because otherwise zooming out
+from a 2x baseline falls straight to 1x and halves the board in one press.
 
-The scale is then CAPPED at 2x. Fitting the panel is not the same as looking right, and past 2x this stops
-reading as a window on a desktop. That is a default, not a verdict — the panel header has a zoom control,
-and `MAX_SCALE` is one constant.
+The BASELINE — the size at 100% — is capped at 2x. Fitting the panel is not the same as looking right, and
+past 2x this stops reading as a window on a desktop.
+
+The panel's own zoom control then multiplies that, so zooming in still goes past 2x and zooming out still
+shrinks. This needs saying because it is not automatic: `fit` is the same number at every zoom level, since
+CSS `zoom` takes off `offsetWidth` exactly what it adds to the pixel ratio. A cap applied to the result
+rather than the baseline therefore flattens the zoom control into doing nothing, which is how the first
+version of it shipped.
 
 Two things the scale has to know about that are easy to miss. `devicePixelRatio` is not the whole story:
 the panel header has its own CSS `zoom`, so the effective ratio is `devicePixelRatio × zoom`, read as
diff --git a/web/MinesweeperPanel.tsx b/web/MinesweeperPanel.tsx
index f64c77a..4cc08c2 100644
--- a/web/MinesweeperPanel.tsx
+++ b/web/MinesweeperPanel.tsx
@@ -28,13 +28,14 @@ const BUTTON_MASK: Record<number, number> = { 0: 1, 1: 2, 2: 4 };
 const MARGIN_DEVICE_PX = 24;
 
 /**
- * The ceiling on how large the game is drawn, whatever room the panel has.
+ * The ceiling on the size at 100%, whatever room the panel has.
  *
  * A taste decision, not a limit of the method: fitting the panel is not the same as looking right, and
- * past 2x this stops reading as a window on a desktop and starts reading as a wall. Raise it and the
- * quarter steps below take over again.
+ * past 2x this stops reading as a window on a desktop and starts reading as a wall.
  *
- * The panel header has a zoom control, so this is a default rather than a verdict.
+ * It caps the BASELINE, not the result. The panel's zoom multiplies what comes out of here, so zooming
+ * in still goes past 2x — the first version of this cap did not make that distinction and flattened the
+ * zoom control into doing nothing, because the fit it capped is the same number at every zoom level.
  */
 const MAX_SCALE = 2;
 
@@ -75,7 +76,7 @@ export const MinesweeperPanel = () => {
   const boxRef = useRef<HTMLDivElement | null>(null);
 
   const [frame, setFrame] = useState<{ width: number; height: number } | null>(null);
-  const [box, setBox] = useState<{ width: number; height: number; dpr: number } | null>(null);
+  const [box, setBox] = useState<{ width: number; height: number; dpr: number; zoom: number } | null>(null);
   const [status, setStatus] = useState<SessionStatus | null>(null);
   const [error, setError] = useState<string | null>(null);
   const [connected, setConnected] = useState(false);
@@ -98,6 +99,7 @@ export const MinesweeperPanel = () => {
         width: node.offsetWidth,
         height: node.offsetHeight,
         dpr: (window.devicePixelRatio || 1) * zoom,
+        zoom,
       });
     });
     observer.observe(node);
@@ -117,14 +119,20 @@ export const MinesweeperPanel = () => {
    * it removes the blur, not the unevenness. A fractional devicePixelRatio produces this at what looks
    * like 1:1, which is the case that is easiest to miss.
    *
-   * QUARTER steps are allowed from 2x up, and that is not a compromise on the above. The unit that has
-   * to stay regular is the CELL, not the pixel: the board is a 16px grid, so at 2.25x a cell is exactly
-   * 36 screen pixels — as even as at 2x or 3x — and only sub-cell pixels alternate. Below 2x that
-   * alternation is too large a share of the pixel, so it stays whole-number there.
+   * QUARTER steps, and that is not a compromise on the above. The unit that has to stay regular is the
+   * CELL, not the pixel: the board is a 16px grid, so every quarter step gives a whole cell — 20px at
+   * 1.25x, 36px at 2.25x — and only sub-cell pixels alternate. Below 1x there is nothing to floor to, so
+   * it takes a unit fraction and the rows that survive are at least evenly spaced.
    *
-   * Quarters rather than halves because the steps are what make the size QUANTISED, and coarse steps
-   * overshoot: with halves this went 2x to 2.5x, from too small to too large with nothing in between.
-   * Finer steps let the picture track the panel rather than lurch between two sizes.
+   * Quarters all the way down rather than whole numbers under 2x, because the zoom control has to have
+   * somewhere to go: with whole numbers below 2x, zooming out from a 2x baseline fell straight to 1x and
+   * halved the board in one press.
+   *
+   * ── zoom multiplies, it does not compete ──
+   *
+   * `fit` is the same number at every zoom level — the panel's CSS `zoom` shrinks `offsetWidth` by
+   * exactly what it adds to the pixel ratio — so zoom has to be applied deliberately or it does nothing
+   * at all. The baseline is what fits (capped), and zoom moves the picture up and down from there.
    */
   const stage = useMemo(() => {
     if (!frame || !box || box.width <= 0 || box.height <= 0) return null;
@@ -133,10 +141,8 @@ export const MinesweeperPanel = () => {
     const usableW = Math.max(1, box.width * box.dpr - MARGIN_DEVICE_PX * 2);
     const usableH = Math.max(1, box.height * box.dpr - MARGIN_DEVICE_PX * 2);
     const fit = Math.min(usableW / frame.width, usableH / frame.height);
-    // Below 1 there is nothing to floor to, so take a unit fraction: the rows that survive are then
-    // evenly spaced instead of dropped in clumps.
-    const snapped = fit < 1 ? 1 / Math.ceil(1 / fit) : fit < 2 ? 1 : Math.floor(fit * 4) / 4;
-    const scale = Math.min(snapped, MAX_SCALE);
+    const wanted = Math.min(fit, MAX_SCALE) * box.zoom;
+    const scale = wanted < 1 ? 1 / Math.ceil(1 / wanted) : Math.floor(wanted * 4) / 4;
     return { width: (frame.width * scale) / box.dpr, height: (frame.height * scale) / box.dpr };
   }, [frame, box]);
 
@@ -297,7 +303,7 @@ export const MinesweeperPanel = () => {
   return (
     <div className="flex size-full flex-col items-center justify-center gap-3 bg-muted/30 p-4">
       {/* min-h-0 on the row, so the canvas gives way to the status line rather than overflowing it. */}
-      <div ref={boxRef} className="flex min-h-0 w-full flex-1 items-center justify-center">
+      <div ref={boxRef} className="flex min-h-0 w-full flex-1 items-center justify-center overflow-hidden">
         <canvas
           ref={canvasRef}
           width={PLACEHOLDER.width}