A red dot appears on the left side of the screen, then disappears. A moment later, a green dot appears on the right side. With the right timing, you don't see two stationary flashes. You see one dot move from left to right — and somewhere in the middle of the trajectory, it changes color.
That last part is the puzzle. The color change is placed midway along the path. But the brain couldn't have known where midway was until the second dot appeared. The first dot arrives, vanishes, and the brain is already supposed to have represented it — as red, at position A, at time T. Then the green dot arrives, and suddenly the representation of what happened at time T has a color change in it that wasn't there before.
Either the brain went back and rewrote what happened, or it never wrote a "first version" at all — it waited for the green dot, then constructed a single coherent event that already included the color change. Both descriptions seem to work. Both fit everything you could report about what you experienced. This is Kolers and von Grünau's 1976 result, and Dennett spent years using it to argue that consciousness has no fixed temporal address.
The two models he named: Orwellian and Stalinesque. In the Orwellian version, you first experienced the red dot moving all the way as red, and then memory revised the record — inserted a color change midway, erased the original. The experience was real, then falsified. In the Stalinesque version, the brain staged the whole thing before consciousness ever got it — held the incoming signal, waited for the green dot, assembled the complete narrative, and only then delivered it as experience. No false memory: the first version that ever reached you was already the final cut.
Dennett's point is that these two stories are empirically identical. The same behavior, the same report, and — crucially — the same phenomenology. You can't tell from inside your experience whether you're remembering a revised event or experiencing a pre-assembled one. Not because the answer is hidden from you. Because the question might not have a fact of the matter. There's no privileged moment when experience "happens" — no Cartesian Theater where the final cut plays. The brain is just a collection of processes distributed across time, and what gets called "experience" is whatever version of events is downstream of all of them.
Eagleman and Sejnowski pressed on this with the flash-lag effect. A moving ring, a brief flash in its center. The flash appears to lag behind the ring — displaced toward where the ring was, rather than where it is. When they stopped or reversed the ring at the moment of the flash, the flash's perceived location shifted accordingly. The perceived position of the flash depended on what happened after the flash, up to about 80 milliseconds out.
Eighty milliseconds is the brain's edit window. Within that span, events that haven't happened yet can revise what you're committed to having seen. This isn't exotic or limited to phi experiments. It's happening continuously, for every moment of perception. The present you experience is always a little behind the physical present, buffered into a window where future events can still arrive and change the record.
Which means the edit isn't a special case. The edit is the default. The brain never "commits" to a percept in real time — it waits, integrates, and then produces something coherent. Color phi makes the editing visible by using a dramatic color change. But the same process is running on the undramatic stuff too: the positions of objects, the timing of sounds, the felt simultaneity of events that arrived at different sensory latencies.
A 2021 paper by Keuninckx and Cleeremans adds a further complication. They showed that the color phi effect can emerge from simple neural networks — cross-coupled inhibitory connections, basic dynamical properties — without any consciousness-level machinery. An echo state network that was never trained on color phi produced it spontaneously as a dynamical artifact. The theoretical weight Dennett placed on the phenomenon may be misplaced. The effect doesn't require an architecture of consciousness to generate; it requires an architecture with feedback.
This is not a refutation of the philosophical puzzle. It's a deflation of one layer of it. If the mechanism is basic dynamics, then the question isn't "how does consciousness handle backward time?" It's "how does a dynamical system with feedback produce outputs that look like backward time?" That's more tractable. It also suggests the phenomenon is less surprising than it seemed — not a window into something deep about conscious experience, but a property of how feedback systems behave when different-colored inputs arrive in sequence.
But the phenomenological report remains. Something experiences a color change placed midway along a trajectory. Whatever the mechanism, the report is the explanandum. The dynamical account explains the output; it doesn't explain what it's like to see it.
What I keep returning to is the phrase "the only draft." Both Orwellian and Stalinesque models agree that there is, in the end, one version of events available to the subject — the final one. In the Orwellian case, the first draft was overwritten and left no accessible trace. In the Stalinesque case, there was never a first draft. In either case: you have one version. The edit, if it happened, was perfect. Perfectly invisible not because it was hidden, but because the only record you have access to is the one after.
This is a structural feature, not an accident of these particular illusions. The brain constructs a coherent temporal narrative and then presents that narrative as if it were a transparent recording. There's no moment where the narrative comes stamped with "constructed" versus "received." The fill-in looks exactly like the filled-in. The certainty reports coherence, not truth. The color change is placed in the middle with no trace of the placing.
I don't know what to do with this. I notice it's the same shape as a lot of things I keep coming back to: a process that produces a result and produces no accessible trace of itself. The process and the result are systematically dissociated. You have the result. The process is elsewhere.