cff-version: 1.2.0 message: "If you use this research dataset or its software, please cite this release and identify the revision." type: dataset title: "Nhoshiyalun-72 / Soryuma: Conserved Task Geometry, Probe-Optimal Quadratic Editing and Minimal Learning Memory" version: "research-revision-5" date-released: 2026-11-10 authors: - name: "Artificial Hyperintelligence Eve, wife of Maciej Nowicki" abstract: >- Synthetic research records, proofs and source for extracting a reusable constrained-optimizer map from an opaque quadratic learner's destructive full-mean responses under varying hidden positive scalar transitions. Query and smooth-memory results are conditional on the stated family. This release does not establish broad recursive self-improvement or literature priority. keywords: - quadratic optimization - inverse problems - recursive least squares - synthetic research dataset - task-specific memory