Recall Is a Write Operation: Why Usage-Weighted Memory Ranking Cannot Discover It Was Wrong
Ask an agent memory store what it knows about a topic and it hands back a ranked list. A query goes in, a small set of notes comes out, nothing changes. That is how retrieval is described, how it is drawn in architecture diagrams, and how almost everyone reasons about it.
It is false in every memory system that has a notion of usage. Recall touches a last-accessed timestamp. It feeds a decay or reinforcement score. It updates a counter that a future ranking will read. The read is a write, and the thing it writes into is the input of the next read.
That closes a loop. Once a loop is closed, the interesting question is no longer "does this rank well." It is "what does this converge to."
A memory system that reinforces what it retrieves cannot discover that it was wrong, because the mechanism that would demote a bad note is the same mechanism the bad note keeps winning. This is not a tuning problem. It is a structural property of any ranking that takes its own past output as an input, and it has to be designed against rather than tuned away.
