diff --git a/sidecar/session.ts b/sidecar/session.ts
index 94032a0..a6ad7e9 100644
--- a/sidecar/session.ts
+++ b/sidecar/session.ts
@@ -318,7 +318,15 @@ async function startInner(log: (line: string) => void): Promise<SessionStatus> {
       // -solid: the stage is bigger than the game, so the surround is now something people look at. Teal
       // rather than the root's black, which reads as a broken stream rather than a desktop. x11vnc sets it
       // itself on a bare display, so this costs no new host binary.
-      ['x11vnc', '-display', display!, '-unixsock', VNC_SOCKET, '-rfbport', '0', '-rfbportv6', '0', '-noipv6', '-solid', STAGE_COLOUR, '-forever', '-shared', '-nopw', '-noxdamage', '-wait', '5', '-defer', '5'],
+      //
+      // -nocursor: the browser already draws a cursor over the canvas, and it is the accurate one. The
+      // client asks for Raw and CopyRect and nothing else — no cursor pseudo-encoding — so x11vnc has
+      // nowhere to send a cursor SHAPE and composites it into the framebuffer instead. That put a second
+      // pointer in the picture, lagging the real one by a round trip and, on a bare X server with no
+      // window manager, drawn as the big black X of the root cursor. Hiding it server-side is the fix;
+      // adding the pseudo-encoding to the client would be a worse one, since the answer is not to draw
+      // the remote cursor better but not to draw it at all.
+      ['x11vnc', '-display', display!, '-unixsock', VNC_SOCKET, '-rfbport', '0', '-rfbportv6', '0', '-noipv6', '-solid', STAGE_COLOUR, '-nocursor', '-forever', '-shared', '-nopw', '-noxdamage', '-wait', '5', '-defer', '5'],
       { DISPLAY: display! },
       'x11vnc',
     );