OES-32 Residual Reference
Abstract
Normative OES-32 residual reference (ADR-001): a deterministic Python definition of the max-absolute residual between two 32-component vectors, with a strict > tolerance failure rule and executable contract tests. Contract. Input: two finite real-valued vectors of length 32 and a finite, non-negative tolerance. Component residual: abs(observed[i] - reference[i]). Aggregate residual: the maximum of the 32 component residuals. Failure occurs if and only if aggregate residual > tolerance. Invalid input is rejected with ValueError. This repository is the normative residual definition for Spark AI NLP under ADR-001 Option A (Residual-canonical); normative commit b77b61254f15778c6ae221843dceac7a8571158e. Downstream oes32_engine (Python) and oes32-hls (C++ HLS) are Profile A sidecars. The OES-512 weighted latch remains TARGET. Evidence tags. Contract-test vectors: SYNTHETIC (hand-written). Empirical, hardware, or field results: none claimed. Scope and limits. Research prototype. It is not an empirical study, a scientific theory, a physical or biological model, a diagnostic, or a claim about consciousness; it is not hardware, medical, or quantum-hardware software, and it does not justify decisions based on a pass/fail result. Evidence tags follow sparkainlp-x/.github (SYNTHETIC / REPORTED / TARGET / UNRUN).
