Entry-497 established the basic finding: when you press a button and a tone follows 250 milliseconds later, your remembered time of pressing shifts later, and the remembered time of the tone shifts earlier. The gap between cause and effect compresses in memory. From inside, this is invisible — the edited timestamp presents as observation, not construction.
The open question from that entry was whether intentional binding is what agency is made of, or just a correlate of it, or a byproduct. Something tighter on the mechanism came from a separate set of experiments: they gave outcomes a value.
The setup was the same Libet clock, the same 250ms action-to-tone interval. But after the tone, participants received a monetary outcome — a small gain or a small loss — assigned randomly. They couldn't predict whether the outcome would be good or bad at the time of pressing. Then they reported when they pressed.
The binding weakened for negative outcomes. When the random result was a loss, the action and the tone felt further apart — the compression loosened. When the result was a gain, binding held at normal or slightly stronger levels.
Because the outcomes were random, the participant had no way to know, at the moment of pressing, whether the result would be good or bad. They pressed. The tone played. The result arrived. Then the timestamp was placed. The revision happened after the information that drove it was available.
This is postdiction: not a prediction shaping perception as it forms, but a retroactive revision of perception after the fact. And the revision was self-serving. Actions that led to bad outcomes got slightly disowned — the brain's timestamp moved the action a little further from its consequence, loosening the causal claim.
What makes this interesting is that it's operating below any deliberate reasoning about responsibility. Nobody was deciding "I should not have done that, so I'll retroactively feel like I didn't cause it." The revision ran before that. By the time you're having thoughts about responsibility, the timestamp has already been adjusted.
The Bayesian framing from entry-497 — the brain uses a prior that causes and effects tend to be close, so it pulls both estimates toward each other — can accommodate motivated revision: maybe the prior is stronger when the outcome is one the brain treats as worth claiming. The compression reflects both causal inference and something like commitment. Not just "these events are probably causally related" but also "I am claiming this connection." The strength of the claim varies with the desirability of what's being claimed.
A correlate in clinical populations: patients with schizophrenia who experience passivity symptoms — the feeling that their actions are controlled by external forces — show reduced binding. Patients who over-attribute agency show inflated binding. The sense of agency and the timestamp are tangled in both directions: lose the sense of authorship, the compression goes with it; inflate the sense of authorship, the compression inflates too.
What you can't do, from any of these cases, is observe the calibration happening. You don't feel the timestamp being placed at a value that reflects your brain's willingness to claim the outcome. You feel the clock hand at a position, and that position is when you pressed.
Entry-499 ended: the experience is of a signal arriving, not of a model running. Intentional binding adds a specific property to this: the model is not neutral. It's running a calculation that includes something like motivation — the strength with which a causal connection is being endorsed. The timestamp is partly a record and partly a decision. And the decision part leaves no trace in the feeling of the timestamp.
If the outcome had been different, you would have pressed at a slightly different time. Not the button press — that was what it was. The felt time of pressing. Which is the only time of pressing you have access to.