How Long a Memory Stays Writable: Self-Revising Records with a Recovering Threshold

Abstract

A companion paper argued that biological memory is a third kind of record, neither kept intact nor erased but retained and rewritten by its own consultation, and gave this a readout-write channel on the record in which a consultation opens the record when prediction error crosses a threshold. That channel has no time in it: the gate opens or does not at the instant of reading. Real records do not behave that way. The reconsolidation window lasts hours, psychedelics reopen a critical period for days to weeks with durations that outlast the drug, and in both cases what made the record writable persists after the event that made it so. We extend the channel by one parameter. After an opening the write threshold is depressed and recovers with a time constant τ; the instantaneous channel is the limit τ→0. Four consequences follow. The record’s writability depends on its own consultation history. Fidelity loss accumulates with the number of consultations inside the window rather than with whether one occurred, under a condition we state, and we give the bound. The index component of the record survives any number of revisions, so the objectivity result of the companion paper holds across a reopened period. The memory sits in a classical gate parameter and not in the environment, which separates host-initiated revision from non-Markovian backflow by a different criterion than before. We instantiate the channel twice, in the Drosophila mushroom body where τ is on the order of hours and in the human self-narrative after a psychedelic session where it is on the order of weeks, and state the observation that would show these are not the same operation. In the fly, τ is measurable now with one added arm to an existing design. For the human case we pose the question the formalism cannot answer and the phenomenology can: which component of the self is index and which is content.