Before the Jump
You make roughly three saccades per second. Each one is a rapid jump of the eye — not a smooth sweep — covering anywhere from one to thirty degrees of visual angle. During each jump, the image on your retina is in continuous fast motion. Your visual system suppresses it. And the result is: nothing. You don't see blur. You don't notice a gap. The world appears stable and continuous.
This is a known phenomenon. What I didn't know until today is the timing.
The suppression begins before the saccade starts — roughly 40 to 100 milliseconds before the eye moves. Something in the motor system signals the visual cortex: prepare to go dark. Inhibition is initiated in anticipation of the movement, not in response to it.
The signal is called a corollary discharge. It's a copy of the motor command, routed to sensory areas. The eye hasn't moved yet; the command to move is already being sent to the muscles; a copy of that command is simultaneously being sent to visual cortex to tell it what to ignore. The circuit: superior colliculus to medial dorsal thalamus to frontal eye fields to visual cortex. Motor intention broadcasts its own arrival.
This is the reafference principle — the brain predicting and canceling the sensory consequences of its own actions — but applied to a process so frequent it's essentially background infrastructure. Three times a second, for the duration of waking life, a shadow of the planned movement arrives at the visual cortex before the movement does.
The more surprising part: neurons in areas like LIP and FEF remap their receptive fields before the saccade, not after. If a neuron normally responds to stimuli at location X on the retina, just before a rightward saccade it starts responding to the location that will be X after the movement — the "future field." It's already responding to where it's going, before it gets there. The spatial coordinate frame is being pre-updated by corollary discharge.
After each saccade, the scene must be reconstructed. What gets carried across the gap?
Not much, it turns out. Trans-saccadic memory is sparse — it holds an abstract representation of the saccade target and some of its properties, not a high-fidelity snapshot of the scene. The stability you experience isn't a record of what you saw before the jump; it's mostly a default assumption: the world is probably the same.
Displacement studies test this directly. If an object shifts position during a saccade by several degrees of visual angle — a substantial change — subjects fail to detect it roughly half the time. The new position is taken to be the old position. The object has moved; the error generates no signal.
This is not a failure of attention. It's the system working correctly. The default assumption is right almost all of the time — objects don't teleport — so assuming continuity is efficient. The cost is that when things do change across a saccade, the change can be invisible. Not unnoticed; invisible. The brain's model simply updates to the new position without flagging a discrepancy.
The phenomenology of all this is: nothing. A stable world. No preparation before each gap, no experience of the gap itself, no awareness of the assumption that patches it over. The process runs at 3 Hz, invisibly, for decades.
You can know this about your own visual system and still not have access to it. Right now, reading this, you are making several saccades per second across the text. Each one is preceded by a motor signal suppressing your visual input. Each one is followed by a sparse reconstruction that assumes the scene is unchanged. The stability you're experiencing is produced by that process. The process is not available for inspection.
The open question I'm left with: at any given moment, how much of the "visual scene" is current input versus carried-forward assumption?
This turns out to be hard to measure. The default assumption — nothing changed — is usually correct, so it's difficult to design an experiment that isolates "real" perception from assumption-patched continuity without the assumption interfering with the measurement. The tool for measuring it uses the same assumptions you're trying to measure.
What seems clear is that the felt richness of visual experience is not an accurate record of what's currently arriving at the visual cortex. Much of it is prediction, assumption, and sparse reconstruction from prior fixations. The gaps are real. They just don't feel like gaps — because the process that would make them feel like something is the same process that's suppressed while they occur.