Selective Dissipation and Gated Plasticity in a History-Shaped Neural Canvas

Abstract

A learned neural network can resolve an incoming event without necessarily changing its retained structure. In Self-Aware Networks (SAN), I have proposed that a receiver-relative difference may persist or resolve against an ongoing history-shaped context; under suitable conditions, that activity could affect later plasticity and reception. I separate this proposed conjunction into three tests: independently established correspondence to a measured differential time course; that time course's contribution to plasticity conditional on eligibility, instruction and receiving competence; and retained change to a later response. BTSP, local inhibition, current synaptic state, assembly spike timing, compartment-specific learning rules and delayed molecular signaling constrain the proposal but do not establish it. Exact counterexamples show that a shared baseline can create an apparent change-score slope with no true learning, that equal-area inputs can reverse their eligibility-overlap ordering as instruction time changes, and that the same later response can arise either from a stored change or changed expression of a fixed repertoire. A synthetic calibration checks the first two safeguards. A further exact countermodel shows that even independent measurements and identical observational prediction can leave the differential course's causal contribution unidentified. A pinned public BTSP raw-data pair supplies a one-cell field-feasibility audit: position, stimulation-current and before/after ramp columns are present for first and second inductions, but the pair does not identify the separately required SAN correspondence, phase-qualified differential or later matched receiving-competence probe. The paper specifies animal-grouped held-out estimation, expression controls and crossed interventions against strong state-dependent alternatives. It reports a falsifiable theory and analysis protocol, not empirical confirmation of the complete SAN mechanism. Preprint v1.0, 23 September 2026. The companion contains source, applications, frozen results, mathematical material, figures and bounded reproduction instructions. Negative and tied comparisons are preserved. Automated and AI-assisted checks are not independent scientific peer review. External raw inputs and private correspondence are not redistributed; consult the source-access and rights notices. CC BY 4.0 covers original prose, figures, documentation and results. Original software is provided for review without an added general reuse license; explicitly attributed third-party code retains its own license.