Where's the weight?
Every picture balances somewhere. Tap the spot where you feel this picture's visual weight settles: not the subject, the mass, the place all its marks, edges and texture would balance on a fingertip. Then see where the instrument puts it.
Tap where the weight balances.
Each dot is one picture: where you tapped, relative to where the lens found the weight (the cross). A cloud leaning one way is a habit of your eye. A cloud gathered toward the frame's center means you pull toward the middle more than the pictures do.
What the lens measures. It finds the mid-strength edges, marks and texture in the picture (the 85th–97th percentile of the gradient field) and takes their center. It knows nothing about faces, objects or meaning. A busy background can outweigh a quiet subject. That gap is the lesson.
Pictures are old paintings (public domain) and AI-generated images. The instrument is the open-source Visual Thinking Lens, kernel canon3-v8, ported to JavaScript; browser numbers are close to, not identical with, the Python reference. Nothing you upload leaves your device.