Answers are grounded in this paper and cite their sources.
Second edition (November 2026). It replaces the first edition (June 2026, doi:10.5281/zenodo.20688568), which reported a self-organised critical transition, a "collapse" at step 788, in a closed-loop Janus memory. That result is withdrawn, and the question behind it is asked with the series' devices: when a loop feeds back strongly enough, like a microphone near its own loudspeaker, it selects a state and holds it. Which state, for how long, and how does it let go? (1) The first edition's collapse. Rerun over 200 noise seeds, the collapse step is 974 ± 311; the map has built-in fixed points at ±1 and falls into one of them with or without noise. It is not self-organised criticality. (2) A fast loop. Two populations of Janus transistors drive each other's gates through a gain stage; the Eu trap of each gate is its own adaptation a. Above a coupling of about 0.71 a held plateau appears. It stays lit while a is below a critical value (0.41 at unit coupling, 0.26–0.56 across gain stages) and otherwise lights and goes out by itself. Cutting the loop removes it within one time constant: the state belongs to the loop, not to either half. The Janus gate has a = 0.07–0.20 and locks. (3) What the plateau holds. In a network of 100+100 transistors storing five patterns, the cue chooses the content, a stronger uncued memory never takes it, and a held loop ignores later cues (0 of 11 taken). Thinner barriers and thicker top oxides (a up to 0.48) make the loop let go, but not yet follow a stream. (4) A slow loop. In the memory loop of Ouroboros, a fresh store tips into obsession above a Ce bias of 0.09–0.12 MV/cm. Healthy fluctuations grow towards that line (detrended variance 5–8-fold). The state is path-dependent: a matured store did not tip up to 0.16 MV/cm. Erasing or resetting the store does not heal it, a reset with abstinence heals for a day and relapses, and lowering the bias heals all seeds. (5) The quantum and observer readings of the first edition are discussed as readings, not physics; lessons for systems that learn from their own output (self-trained models, recommenders, replay) are offered as hypotheses. Files: paper in English, Traditional Chinese and Japanese (the English version governs); a primer in each language; code and sources (code_and_sources.zip) reproducing every number and figure, including the first edition's code for the rerun. Computational throughout; nothing is measured. Part of the Angelica Paper Series. Licence: text and figures CC BY 4.0; code Apache-2.0. Defensive publication: disclosed as prior art, free for anyone to use and not to be patented by others. This edition replaces the first edition's licences (CC BY-NC-ND 4.0; BUSL 1.1). Keywords: attractor dynamics; persistent activity; working memory; thalamocortical loop; ignition; adaptation; saddle-node bifurcation; critical transition; early-warning signals; hysteresis; memory consolidation; model collapse; charge-trap synapse; europium; GaN HEMT; retraction; Angelica Paper Series; defensive publication
No takes yet. Share an insight, caveat, or question.
Tung Ning Liu (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: