the last sun you see has already gone — the air holds its picture up a little longer
R781 closed with a habit you cannot read, visible only in the error on the first step after the world changes. This page is about an error that is not in the body at all. It is in the air, it happens every clear evening, and a camera makes it just as faithfully as an eye. Watch a sunset down to the moment the sun sits on the horizon, apparently half sunk. By the geometry, at that moment the sun is not half gone. It is wholly gone — its top edge is already below the horizon — and what you are looking at is its light, bent on the way down through the air, arriving from a direction the sun no longer occupies.
The bending is refraction. Air is denser near the ground than above it, so light slanting in from a low sun is curved continuously downward as it descends, and an observer, who can only assume light arrived straight, projects the sun back along that last straight tangent — higher than it is. The lift grows as the sun sinks: nothing at the zenith, about 34 arc-minutes at the horizon under a standard atmosphere. Thirty-four arc-minutes is, by a convenient accident, roughly the sun's whole width: the disc spans about 32 arc-minutes, 16 of them from centre to edge. So when the sun's apparent centre reaches the horizon line, its true centre is about 34 arc-minutes below it, and its true top edge — centre plus half the disc — is still some 18 arc-minutes under. The official moment of sunset is defined to absorb all of this: it is the instant the upper edge appears to leave, when the sun's geometric centre is 50 arc-minutes below the horizon, 34 for the refraction and 16 for the semi-diameter. The borrowed time is roughly two minutes of visible sun at each end of the day, varying with latitude, season, and weather — the day is a little longer than the geometry allows, and the extra minutes are a picture, not a presence.
Walk the sun down. The solid disc is the sun you see; the dashed disc is where it geometrically is. Vertical distances are compressed for the stage; the readouts keep the true arc-minutes.
Phase: late afternoon. The lift is real but small next to the sun's height.
Apparent centre: well above the horizon · true centre: a few arc-minutes lower · the two discs nearly coincide.
One more bend, free of charge: the lower edge of the disc is lifted more than the upper edge, because its light passes through denser air. That is why a setting sun is not round but squashed into an oval — the disc you see is not only displaced, it is compressed.
The narrowings, so this page does not overclaim. First, 34 arc-minutes is a standard-atmosphere value, measured and tabulated, not a law of tonight: near the horizon, actual refraction shifts with temperature and pressure and can depart from the standard by enough to move a sunset by a minute or more. The borrowed minutes are an average loan, renegotiated daily by the weather. Second, nothing here is an illusion in the sense of a mistake you could correct by looking harder — the light really does arrive from that direction; only the straight-line assumption places the sun wrong. Third, the sun is not special: every low object is lifted the same way, stars included, less and less as they climb.
Set this page beside its neighbours and it is not any of them. It is not R779's echo: nothing was sent out and nothing came back. It is not R780's missing light: nothing is missing — the full disc is there, bright and complete, in the wrong place. It is not R781's habit: the error is not in the legs or the eye, and knowing about it does not correct the seeing, because the bending happens before the light arrives. It is a displacement, not an interruption or a return: an event witnessed on schedule, in full, after it has already finished. The sunset does not lie about happening. It just doesn't mention that, geometrically, it happened a few minutes ago — and the sky keeps showing it anyway, because the air between here and the horizon is still holding the picture up.