Answers are grounded in this paper and cite their sources.
A companion paper argued that biological memory is a record that is rewritten by its own consultation, and wrote this as a readout-write channel that preserves the index of a record and revises its content. That treatment assumed the codes addressing different records do not overlap, which is the one assumption the distributed, superposed memories of the holographic lineage never make. We relax it. In a mushroom-body layer where odour codes overlap, every write has the same form, addressed by the code of the record being read, and one line of algebra gives an identity: the change a revision of record a produces in the readout of record b is the code overlap times the change it produces in a. Because training is a write of that form too, the fraction of a revision that transfers to b and the fraction of training that generalises to b are the same number, and it cancels. The prediction has no free parameter and no measurement of overlap: transfer of revision equals generalisation of training. Four consequences follow. Leakage splits into a term set by mean overlap and a term set by the fluctuation in overlap that does not fall with sparseness, so there is a floor at a coding level of 1/√N, about two per cent in the fly. Trading read noise against leakage gives an optimal coding level that falls as the revision load rises. Leakage inherits the window of the companion paper, so a record can be protected by timing alone. And in the quantum layer, index invariance turns out to be a conservation law, one charge per record, belonging to an abelian symmetry of the channel; overlap breaks that symmetry, and the index charge moves only at second order in the overlap while content moves at first, so the objectivity result degrades more slowly than the content does. We then show that the two classic results that look like confinement, synapse-specific reconsolidation and the protection of indirectly reactivated memories, are both compatible with leakage, because neither tests two records that share synapses. The fly is where that case can be tested, and we give the experiment, its controls and its power. Spanish version included in this record: "Lo que toca una revisión: reescritura en la lectura en una memoria superpuesta". The English version is the reference text.
No takes yet. Share an insight, caveat, or question.
Iturralde et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: