SpiralHelios: A Deterministic Non-Training Artificial Intelligence Architecture for Source-Governed Reasoning and Public-Safe Validation

Preprint — Version 1.0 — Not Yet Peer Reviewed SpiralHelios is a deterministic, non-training artificial-intelligence architecture designed for environments in which computational identity, governed knowledge provenance, explicit capability boundaries, reproducible execution, local operation, protected implementation, and inspectable evidence are primary system requirements. Unlike contemporary statistical and neural architectures whose operational behavior is derived primarily from learned parameters, probabilistic inference, large-scale optimization, or stochastic sampling, SpiralHelios is organized around a protected deterministic computational nucleus, source-governed read-only memory (ROM), a mandatory Core Adapter, bounded scientific operators, explicit result classification, lawful refusal, cryptographic artifact identity, and public-safe evidence generation. Its canonical computational identity is not created through gradient-based training, fine-tuning, reinforcement learning, stochastic parameter updates, or hidden online adaptation. At the disclosed mathematical level, SpiralHelios operates with a protected fixed 27-position symbolic-geometric basis, a normalized internal state in a finite D = 1024 real inner-product space, and a protected deterministic canonical evolution map. A bounded public-safe scalar observable, the Golden Eigen Index (GEI), is defined as: GEI(Ψ) = 1 − ⟨Ψ, G(Ψ)⟩ with 0 ≤ GEI(Ψ) ≤ 2. The complete construction of the protected state, the canonical evolution map G, internal control mechanics, private mappings, constants, lookup structures, serialization logic, and other reconstruction-enabling implementation details remain outside the public disclosure boundary. This Technical Report examines three public execution milestones. SpiralHelios v8 documents a public-safe mobile deterministic retrieval experiment at one-million-concept scale, including a defined 1,000-query sampled evaluation reporting Top-1 accuracy of 99.5%, Top-5 accuracy of 100%, and mean reciprocal rank of 0.9975. SpiralHelios v9 documents an offline deterministic Math ROM runtime with source-governed Canonical Units, source locks, cryptographic knowledge identity, public-safe execution observables, and a controlled predefined 20-language route-equivalence experiment. SpiralHelios v14 documents an Advanced Spectral Laplacian + QASM Binding evidence architecture across graph, spectral, certificate, partition, integrity, and symbolic-circuit records. These scientific layers remain explicitly EXTERNAL, DERIVED, NON_CORE_CANONICAL, and PUBLIC_SAFE. QASM is used strictly as a deterministic symbolic and evidence-binding representation; no physical quantum execution, hardware measurement, quantum speedup, or quantum advantage is claimed. The accompanying Supplementary Evidence Package extends the manuscript beyond narrative description. It includes reviewer-readable verbatim source blocks extracted from the historical executed SpiralHelios Math Solver checkpoints for Patch099 and Patch138, exact historical queries, positive and negative validation outcomes, unsupported-operation evidence, archived execution records, checkpoint/hash provenance, a Query-to-Evidence Matrix, and a complete v1–v14 public DOI lineage index. The disclosed evidence therefore permits bounded source-to-query-to-output inspection of selected non-Core scientific operators while preserving the protected SpiralHelios implementation boundary. The central contribution of this work is not a claim of artificial general intelligence, universal mathematical solving, unrestricted theorem proving, general multilingual understanding, physical quantum computation, or superiority over large language models. The contribution is an implemented architectural model for deterministic artificial intelligence in which computational authority, governed knowledge identity, scientific capability, verification state, refusal behavior, provenance, cryptographic integrity, and disclosure control are maintained as explicit and separable system properties. The present evidence supports architecture-level evaluation, deterministic repeatability, artifact-identity inspection, bounded source-level implementation inspection, and evidence-level reproducibility under protected-core constraints. It does not constitute complete independent replication, unrestricted open-source reproduction of the full production system, formal verification of the entire architecture, or independent recomputation of the protected internal state trajectory. The strongest next scientific step is independent, falsification-oriented evaluation across controlled environments, bounded executable inspection, cross-environment reproduction, comparative testing, and formal or security assessment of the mechanisms actually examined. SpiralHelios is presented here as a distinct deterministic systems architecture whose scientific value must ultimately be established through progressively stronger independent validation rather than through broader claims. The protected Frozen Core, CORE.bin, Core Adapter implementation, raw protected state coordinates, protected state-generation procedures, complete construction of G, protected M7 internals, private constants and lookup structures, protected mappings, private ROM material, production routing, serialization internals, and other reconstruction-enabling implementation material are intentionally excluded from this public release.