NHOS Oracle Matrix Search Engine™ A Local Hybrid Retrieval and Contextual Discovery Architecture for Personal Health Knowledge

Abstract

NHOS Oracle Matrix Search Engine™: A Local Hybrid Retrieval and Contextual Discovery Architecture for Personal Health Knowledge Technical Research White Paper — Version 1.2 · September 2026Corresponding Software Version: NHOS Oracle Matrix Search Engine v8.0.0Research Stage: Implemented Research System & Validation Framework NHOS Labs, Inc. — NHOS Intelligence Matrix Engine™ Research Program Overview This technical research white paper presents the architecture, retrieval methodology, implementation model, and planned empirical validation framework of the NHOS Oracle Matrix Search Engine™, a local hybrid retrieval and contextual discovery subsystem developed within the NHOS Intelligence Matrix Engine™ (NIME™) Research Program. Oracle is designed as a structured "front door" to NHOS health knowledge. Rather than relying on a single retrieval mechanism, the engine combines exact and content matching, title and excerpt weighting, token-level matching, Levenshtein-based fuzzy matching, category weighting, type-specific ranking, multi-concept query processing, and contextual discovery. The architecture is deliberately non-semantic and non-ML-based. Core retrieval does not depend on embeddings, vector databases, large language models, or remote inference. The system is designed for local, offline-capable execution, supporting privacy-conscious health-information discovery while minimizing dependence on external services. Research Contribution The Oracle research program investigates whether a deterministic, hybrid retrieval architecture can provide effective and reproducible retrieval across heterogeneous personal-health knowledge while retaining local execution and explicit system provenance. The research emphasis is therefore not simply on implementing a search interface, but on experimentally evaluating the contribution of individual retrieval mechanisms and their combined effect on retrieval quality. The validation framework is designed to investigate: retrieval relevance; precision and recall; ranking quality using MRR and nDCG; confidence behavior and calibration; contribution of individual retrieval mechanisms through ablation studies; comparison with conventional retrieval baselines; query-stratum performance; latency and computational behavior; scalability; retrieval error patterns; privacy and network behavior; voice-query equivalence; and reproducibility across documented software, corpus, configuration, and runtime versions. Implemented Architecture Version 1.2 documents the evolution of Oracle from software v6.3 through v8.0.0, including the following implemented capabilities: Configuration-driven retrieval.Scoring constants, routing triggers, and runtime limits are exposed through a versioned ORACLE_CONFIG structure, supporting reproducible configuration and controlled experimentation. Hybrid retrieval.The engine combines exact/content matching, token matching, Levenshtein-based fuzzy matching, title/excerpt weighting, category weighting, and type-specific ranking. Multi-concept processing.Queries containing multiple concepts can be decomposed and processed across the heterogeneous NHOS knowledge corpus. Deterministic recommendation mode.Hash-seeded candidate selection replaces uncontrolled randomization, enabling reproducible recommendation experiments. Parameter-free rank-based confidence.The default confidence method uses rank-based confidence rather than assuming that raw retrieval scores are directly comparable across queries. Runtime provenance.Engine, schema, corpus, and configuration metadata are exposed through the public JavaScript API, including NHOSOracleHybridSearch.getStats(). Embedded regression testing.The v8.0.0 distribution includes ten executable pure-function assertions covering core retrieval behavior; the shipped release reports 10/10 tests passed. Corpus integrity verification.External corpus files can be verified against recorded SHA-256 hashes before ingestion. HL7 FHIR R4 export.The engine provides a FHIR R4 Bundle export with resource-specific attributes and provenance metadata. Internationalization.The interface includes English, Spanish, French, and German language dictionaries using a data-driven internationalization mechanism. Voice retrieval interface.Browser-native Web Speech API integration supports voice-query input with interim transcript handling and language-aware processing. Strict-CSP modular distribution.The engine is distributed both as a self-contained single-file implementation and as a modular HTML/CSS/JavaScript implementation compatible with strict Content-Security-Policy environments. Validation Status The Oracle system described in this publication is an implemented research system accompanied by a defined validation framework. The white paper does not claim that Oracle has already demonstrated superior retrieval effectiveness, superior recommendation quality, calibrated confidence, clinical effectiveness, superior speech-recognition accuracy, or improved health outcomes. Those claims require empirical investigation. The planned evaluation, scheduled for Q4 2026–Q2 2027, will use controlled benchmark queries, predefined relevance judgments, baseline comparisons, ablation experiments, latency measurements, and reproducible execution conditions. This distinction between implemented capability and empirically demonstrated performance is intentional and central to the research program. Software Correspondence Paper version: 1.2Oracle Engine: 8.0.0Knowledge Schema: 1.0.0Corpus: 1.0.0Confidence Method: rank-basedInterface Languages: English, Spanish, French, GermanBuild Date: 11 September 2026 Runtime provenance is available through: NHOSOracleHybridSearch.getStats() The embedded regression harness is available through: NHOSOracleHybridSearch.runTests() The v8.0.0 release reports 10/10 regression assertions passed. Companion Software The corresponding v8.0.0 Oracle engine is deposited separately as companion software. Distribution entry points include: oracle_upgraded.html — self-contained single-file implementation oracle/oracle_upgraded.html oracle/oracle.css oracle/oracle.js — modular strict-CSP distribution Core retrieval requires no server, build system, or network connection. Research and Collaboration Use This publication is intended as a technical reference for researchers, engineers, health-informatics investigators, clinical and institutional collaborators, technology evaluators, and prospective research partners interested in: health-information retrieval; local-first computing; privacy-conscious health technology; hybrid lexical retrieval; fuzzy and multi-concept search; structured health knowledge; provenance-aware system design; reproducible retrieval experiments; and evaluation of deterministic intelligence architectures. The system is positioned as a research platform from which controlled empirical studies can be conducted. Future evaluation results will be reported separately from this architectural publication. Research Principle Build transparently. Measure rigorously. Report proportionally. NHOS Labs, Inc.NHOS Intelligence Matrix Engine™ Research ProgramNHOS Oracle Matrix Search Engine™Technical Research White Paper v1.2 · September 2026 Keywords health information retrieval; hybrid retrieval; fuzzy matching; Levenshtein distance; local-first software; personal health knowledge; knowledge discovery; domain-aware ranking; multi-concept search; contextual discovery; privacy-oriented architecture; NIME™; NHOS Oracle™; configuration-driven retrieval; FHIR R4; voice search; Web Speech API; internationalization; reproducible research; health informatics License Paper text: Creative Commons Attribution 4.0 International (CC BY 4.0)