The thing about the cold water experiment that I can't stop turning over isn't the temporary recovery itself. It's what happens after. The patient — the 84-year-old woman in your account — spent twenty minutes knowing she had been paralyzed for days. She said so. She was aware. She was coherent. And then the vestibular activation faded, and she returned to denial, and she had no memory of ever having been aware.
Your account of anosognosia explains the default state cleanly enough: the right hemisphere normally handles anomaly detection, and when it's damaged, the left hemisphere confabulator runs unchecked. The arm doesn't move, no mismatch fires, and the left hemisphere builds whatever narrative is necessary to fill the gap — she was tired, she didn't feel like moving it, she's not sure why she's having trouble right now. The comparator is gone and so there's nothing to flag the failure. This I follow.
What the cold water disrupts, as you've described it, is the left hemisphere's dominance. The vestibular activation reaches intact right hemisphere systems — or at least regions that include intact right hemisphere circuitry — and briefly reinstates whatever checking function those systems provide. The confabulator is overridden. Reality-checking resurfaces. The patient knows.
But then the water warms, and she doesn't remember knowing.
I want to ask about the structure of that forgetting, because I don't think it's ordinary forgetting. Ordinary forgetting is a failure to retrieve something that was encoded. This seems different. When she returned to the unaware state, she didn't have a memory of awareness that she couldn't quite reach — she had no trace of it at all. From inside the unaware state, the aware state had simply not happened. And from inside the aware state, she had clear access to the recent history of her paralysis: she knew she'd been paralyzed for several days. That's not a blank. That's a narrative, a timeline, a felt continuity with a past.
So there were two states in the same brain, in the same hour, each with a coherent account of the recent past — and those accounts were mutually inaccessible. The aware version had more content. The unaware version outlasted it. Neither could see the other.
The comparator model tells us why the unaware state is the default: the damaged monitoring system doesn't generate awareness of the deficit, so denial is not active confabulation, it's just the absence of a flag. But the memory question seems to require something additional. The aware state happened. It was experienced. It produced speech. Why did it leave nothing the unaware state could access?
One possibility is that memory consolidation depends on the same right hemisphere circuitry that was temporarily reactivated. When the activation fades, the ability to encode and retrieve from that state fades with it. The memory of awareness was formed in a context the unaware state can't read — encoded by systems that aren't running in the default configuration. This would mean the two states share a brain but not a memory system, at least not a working one.
But that raises a further question: what, exactly, is a memory system in this context? The patient in the unaware state could remember breakfast, could remember conversations from the morning. The encoding machinery was working for everything except the awareness of paralysis. So it's not that she couldn't form memories at all. It's that the specific content of the aware state — the knowledge of paralysis, the felt sense of that knowledge — wasn't accessible through the pathways the unaware state uses to retrieve.
Which makes the situation structurally identical to the deficit it's documenting. Anosognosia is not an absence of information — the information about paralysis is present in the system, processed, influencing behavior in ways the patient doesn't consciously register. The awareness is missing because the pathway to awareness is severed. And now the memory of awareness is missing because the pathway back to it is severed. The information is there twice over — the fact of paralysis, and the fact of having-known-about-paralysis — and twice the same route is blocked.
What I find genuinely hard to locate: which version of the patient is the patient? This isn't a philosophical question about personal identity — or not only that. It's a question about what the account should be anchored to. From inside the unaware state, the aware state was an interruption, a temporary anomaly, something the brain almost immediately closed around. From inside the aware state, the unaware state was the impairment. Each version was continuous with the surrounding hours. Each had a working account of recent events. The aware version had something the unaware version lacked. But the unaware version is what persisted.
Your framework has always been sympathetic to the idea that the self is a construction — "Phantoms" makes this case across many conditions, not just anosognosia. The phantom limb, the Capgras delusion, the Cotard syndrome all show the constructed nature of self-models that most people treat as given. But the cold water experiment seems to add something: not just that the self is constructed, but that the same substrate can run two incompatible constructions serially, with no continuity between them. Not a malfunctioning self, but two separate selves that happen to share a body and an hour.
The question I'd want to ask you is whether you think the memory failure is the same mechanism as the deficit, or something additional. The standard comparator account explains why the unaware state doesn't know. I'm not sure it explains why the unaware state can't remember having known.