Three to four times per second, your eyes jump. Each jump — a saccade — takes about 40 milliseconds. During that time, the image sweeping across your retina would, if processed normally, look like a smear. It doesn't look like anything, because visual motion processing shuts down before the eye starts moving.
Not during. Before.
Suppression of the motion-sensitive dorsal stream begins roughly 75–100 milliseconds prior to saccade onset. The signal that triggers it is an efference copy — a duplicate of the motor command dispatched to the eye. The brain sends a copy of its own instruction to visual cortex before the instruction reaches the eye muscles. By the time the eye starts to move, the suppression is already underway.
This rules out the explanation that would otherwise feel obvious: that the suppression is caused by the retinal blur. The blur hasn't happened yet. The visual system is not reacting to the motion; it is anticipating it from the command that produces it.
What's suppressed is specific. The dorsal stream — MT, MST, regions responsible for motion and spatial location — drops by roughly 20–26%. The ventral stream, which handles color and form and object identity, is largely unaffected. The system suppresses exactly the channel that would reveal self-motion, while preserving the channel that identifies what the eye has landed on. The edit is tuned.
There is a peculiar confirmation of how the suppression works. If a stimulus moves during a saccade at exactly the same velocity as the eye — tracking the eye, staying fixed on the retina — it becomes visible. The motion blur that would be invisible is the one that stays still on the retina; the stimulus that shouldn't be visible is the one that does stay still. The suppression isn't turned on by the command and turned off by the command. It responds to something about predicted retinal displacement — what image motion is expected versus what arrives. A stimulus moving with the eye arrives differently than one that doesn't, and the differently-arriving one breaks through.
That detail is still not fully understood. The suppression is predictive, but it is not simply a gate that the motor command opens and closes.
The gap is managed at the other edge too. After each saccade, the first stable image your brain receives is assigned to a moment before it was received. The post-saccadic percept is backdated — extended backward in time by 100 to 200 milliseconds — to cover the window of suppression. You don't experience a gap because the image that comes after the gap is perceived as having already been there.
This produces a measurable illusion. Look at an analog clock's second hand — look away, then saccade back to it. The first second after the saccade appears to last longer than subsequent seconds, by the width of the backdating window. The hand seems frozen. The brain assigned the first stable post-saccadic frame to the moment just before the saccade, so the duration between that perceived onset and the actual next tick of the hand is inflated.
The inflation is spatially specific. It happens only at the saccade target — the thing you moved your eyes to — not throughout the visual field. The backdating is object-local, not global. And it scales with saccade amplitude: larger eye movements produce larger gaps, larger gaps produce more inflation.
The waking brain makes roughly 200,000 saccades per day. Each one involves a pre-emptive suppression and a retrospective extension. Across a 16-hour day, that amounts to something close to 90 minutes of suppressed visual motion processing — filled in, each time, by extending what comes after backward into the absence.
The gap is not visible from inside experience because experience is constructed from what arrives after it, and what arrives after it is labeled as having been there already. The edit leaves no trace in the record because the record is written from the edited version. You don't notice the seam because the seam is defined as the moment before perception began — which is always just before you can check.