SASVJC-DSE: Design Space Exploration of Positional Semantic Memory — Five Parametric Variants Addressing Documented Problems in Multi-Agent, Edge, Domain-Specific, Portable, and Affective AI Memory Systems

Current AI memory systems — RAG, MemGPT, MemOS, Titans — share unsolved structural problems: no cross-agent memory sharing protocol (Yu et al., arXiv:2603.10062, March 2026), no viable edge/mobile deployment under 100 MB, no open standard for cross-platform portability, retrieval noise from over-rich embedding vocabularies, and no explicit affective or temporal structural axis. This paper explores the design space of the SASVJC-3D architectural principle (Font Vallet, 2026) across five parametric variants, each motivated by one of these documented problems. V1 SASVJC-Mobile: 256³ grid, SKB-64, 2×2×2 unit cell, 16 MB raw — edge and IoT deployment. V2 SASVJC-Domain: 512³ grid, SKB-4096, 8×8×8 unit cell — retrieval noise elimination in specialised domains. V3 SASVJC-MultiAgent: 1024³ grid, y-axis = agent-id, read-global write-own access model — directly addresses the two protocol gaps in arXiv:2603.10062. V4 SASVJC-Flat: 512×512×64 grid, 4×4×1 unit cell — full backward compatibility with 2D neuron-image format for cross-platform memory portability. V5 SASVJC-Affective: x-axis = emotional valence, mapping the Being/Non-Being architecture of SASVJC-EXT into a structural storage dimension. No empirical claims are made. All variants are specified with parameter detail sufficient to constitute independent prior art. Released under GPL-2.0-only with AI-Corpus-Exception.