A rat gets a mild footshock — mild enough to leave an impression, not strong enough to form long-term memory. Next day, the rat doesn't remember the context. Same rat, same shock, but twenty minutes earlier it explored a novel environment for five minutes. Next day, it remembers.
The footshock was identical. The rat's response to it was the same. What changed was what happened nearby in time.
Frey and Morris described the mechanism in 1997. A weak event — the footshock, or weak electrical stimulation at a synapse — sets a tag. The tag lasts roughly an hour. It isn't strong enough to trigger protein synthesis on its own, so the synapse stays in a kind of provisional state. But if during that hour a strong event occurs in the same neuron, the plasticity-related proteins synthesized by the strong event can be captured by the tagged synapse. The weak memory gets converted to long-term storage.
The key word in their framing is "commitment." What weak stimulation creates is "the potential for lasting change, but not the commitment to such a change." The tag is the potential. The proteins are what make it real. And the proteins come from somewhere else — from whatever strong event happened to occur in the adjacent time window, in overlapping neural circuitry.
The synapse that encoded the weak event has no signal about which way this goes. The tag is either captured or it fades. There's no announcement either way.
At the behavioral scale this extends to something stranger. Dopamine and norepinephrine released during novel or surprising experiences trigger hippocampal protein synthesis. Memories being formed during that window — even for entirely unrelated content — become eligible to capture those proteins if their tags are still active.
Researchers ran the experiment with school children in Argentina. Children heard a story. Thirty to sixty minutes later, they attended a novel music lesson. The next day, memory for the story was improved. Novel music lesson four hours after the story: no effect. The proteins had dispersed.
The story was the same. Attention was the same. Whether it's still there the next morning depends partly on whether something exciting happened in the following hour.
Competition makes this more complicated. Multiple weak events can be tagged at the same time, but there's a limited supply of proteins. When they arrive, they don't preferentially go to the most important memory. They go to tagged synapses — and the first one captured may use what the second one needed. Memory doesn't only compete with forgetting. It competes with adjacent memory for the same molecular resource.
In one experiment, rats underwent weak training for two separate tasks within the effective time window. Memory for one improved significantly while memory for the other deteriorated. The "winner" wasn't determined by which experience was more important. It was determined by timing and the distribution of synaptic tags.
One more finding worth sitting with. Participants who benefited from retroactive memory enhancement showed higher rates of source misattribution: they were more likely to think they'd encountered items at the time of the salient event, not the actual time. The mechanism that strengthened the memory also introduced errors about when the memory formed — errors along the exact dimension the mechanism works along.
So it isn't just that the persistence of a memory depends on context the synapse can't observe. The mechanism that determines persistence leaves a fingerprint in the content of what gets stored.
The ordinary account of memory is that important things get remembered. Importance determines encoding strength, encoding strength determines durability. There's a rough truth in this. But it's incomplete in a specific way.
Whether an experience is still accessible tomorrow depends partly on what happened in the following hour — not what happened in the experience, but what happened near it, in adjacent neural populations, in the window before the tag fades. The event that gets preserved and the event that provides the proteins for preservation can be completely unrelated. Their connection is temporal and anatomical, not semantic.
From inside the encoding, there's no way to distinguish a memory that will persist from one that won't. They feel the same while they're forming. The synapse can't see what happens next, and the proteins, if they arrive, come without explanation. The gap between "something was encoded" and "something will be remembered" is invisible from within the encoding event itself.