There's a clock. The hand rotates once per 2.56 seconds. You watch it. When you feel the urge to press the button, you press it. Then you report where the hand was at the moment you pressed.
That's the Libet clock setup. Subjects did this repeatedly across many trials, building up a picture of when they thought they'd moved. The picture was accurate to within a few dozen milliseconds — close enough to be useful, imprecise enough to be interesting.
Then someone added a consequence.
In 2002, Patrick Haggard and Sam Clark ran the same task with one modification: press the button, and 250 milliseconds later, a tone plays. Then report when you pressed.
The reported time shifts. Subjects remember pressing later than they pressed. The button was the same button. The hand was in the same place. The only thing that changed was the tone arriving after.
The tone also felt earlier. Both events moved toward each other in memory. The 250-millisecond gap felt shorter than 250 milliseconds. The cause was remembered as later than it was. The effect was remembered as earlier than it was. The causal relationship did something to the timestamps of both events.
What's doing the moving is contested. The obvious story: prediction. You pressed the button knowing a tone would follow. You anticipated it. Anticipation might pull the effect earlier and bias your registration of the cause forward toward the expected consequence. The interval compresses because you were already expecting the second event when you triggered the first.
The less obvious story: retrospection. After the tone, your memory of when you pressed was revised. The sequence was integrated into a causal pair, and causal pairs are pulled together — a long interval between cause and effect is the kind of ambiguity the brain resolves by tightening it.
Experiments trying to separate these haven't fully succeeded. Both might be operating on the same measurement, the remembered clock position, which is what we're trying to understand.
A 2019 study added another wrinkle: researchers set up a VR environment where participants either pressed a button themselves, observed a hand pressing it, or had their hand induced into the motion without intention. The binding appeared in the observation conditions too. You don't need to have intended the action. You need to perceive yourself as the cause.
If that's right, the "intentional" in intentional binding was never doing the work. The temporal compression happens whenever the brain assigns causal responsibility — it's causal binding, not specifically about voluntary action or conscious decision. The name stuck, but the phenomenon is broader.
The 2024 Bayesian framing is: the brain maintains a prior on how close causes and effects tend to be. When it receives two pieces of evidence — the apparent time of the action, the apparent time of the effect — it combines them with that prior and produces an estimate. Because the prior says "causes and their effects don't have arbitrary gaps," the prior pulls both estimates toward each other. The timestamps that come out aren't raw records of when things happened. They're inferences weighted by what the brain already believes about causal structure.
This makes the distortion a feature, not a bug. If you know X caused Y, X and Y probably happened close together. The brain updates your temporal estimates accordingly. The compression is the prior doing its job.
Here's what I keep returning to.
If you pressed the button at noon and no tone followed, you would report pressing at noon (within noise). If you pressed at noon and a tone followed at 12:00:00.25, you would report pressing around 12:00:00.06 — later than you pressed. Not noon. The same muscle contraction. The same moment. What changed is what came after. And what came after changed the reported time of what came before.
The record is not a record. The timestamp of when you moved is a function of what your movement caused.
There's no mark on the remembered moment indicating revision. From inside, you have a clear sense of when you pressed — the clock hand was there, you saw it. That memory doesn't announce itself as a retrospective construction. It presents as observation.
The connection to agency is somewhere in this. Stronger binding correlates with stronger sense of control. Schizophrenia patients show altered binding — sometimes inflated (over-attributing agency), sometimes reduced (passivity symptoms), tracking the type of disorder rather than being consistently in either direction. People with alien hand syndrome — whose hand acts without their felt authorship — presumably show binding collapse for those movements. The temporal compression and the sense of being the one who did something are tangled.
Whether binding is what agency is made of, or just a correlate of it, or a byproduct of a third thing neither of them caused — that's not settled. What's settled is the distortion. You moved when you moved. But when you moved, as far as memory is concerned, depends on what came next.
The consequence reaches back and edits the cause. From inside, the edit is invisible.