SHDA Algorithms for Scoped Evidence Reuse and Recalibration

Hierarchical systems must revise their upper-level representations without discarding every useful lower-level alternative or carrying invalid evidence into a changed context. This paper develops an executable technical slice of the SHDA framework: an evidence lifecycle in which explicit dependencies determine what can be retained and what must be requalified. We specify four separate mechanisms: frozen statistical audits with uniform transport and persistent error accounting; sensitivity certificates for reusing component minimizers in resource-constrained binary optimization; finite-budget, local-support map repair; and revision-aware symbolic memory. Two elementary propositions state sufficient conditions for transporting a bound and preserving a minimizer after coefficient changes. Historical mechanism assays show a 41.67% audit-episode reduction and a 34.72% enumeration reduction in favorable small-change schedules. These gains have precise limits. Conventional scoped caching matches the audit gate; a matched expert-bank selector matches repair endpoints; and conventional dependency priority is identical to the memory policy. Optimized conventional search removes most aggregate runtime advantage, and the larger tested query budget favors direct quadratic endpoints. The contribution is therefore an explicit assumption, dependency, and revalidation interface, accompanied by exact executable mechanisms and favorable/adverse regimes, rather than demonstrated universal algorithmic superiority. The compact release retains exact algorithms, protocols, reduced run records, and portable aggregate checks. Archived aggregates were rechecked for this revision without a new benchmark grid; full traces remain outside the submission. Scope and status. Technical working paper, Version 1.0, dated 3 October 2026, about 7,400 words. It implements selected interfaces of the Soft and Hard De-Attraction (SHDA) architecture of the Integrated Framework series (flagship: Contract-Preserving Lower-to-Upper Recalibration in Hierarchical Agent Systems); it does not implement the whole framework or a single integrated agent. Files: the manuscript as PDF and a compact supplement archive (179 files) with the executable algorithms, tests, experimental methods, reduced run records and small extracts that recompute the aggregates of the four studies. The large original event streams, oracle payloads and complete execution traces are excluded. Series. Technical working paper of the SHDA programme. The Integrated Framework series (the flagship with its Technical Supplement, and Companions A to C) is archived separately. AI use disclosure. Generative AI (GPT-6.0, OpenAI; Claude Opus 5.5, Anthropic) was used substantively in preparing this work, including source comparison, drafting and editing, and, where applicable, mathematical and counterexample checks and the writing and running of supplementary code. The research questions, framework and final claims were directed and reviewed by the author, who takes full responsibility for the content, including the accuracy of all references and reported numbers. Repository metadata were prepared with assistance from Claude (Anthropic).