An independent reference implementation of the mechanisms described in Executable Trust: The Runtime Architecture of Production-Ready Enterprise AI. The artifact implements a versioned trust contract with stable rule identifiers, a pure fail-closed decision function over claim-level verdicts, evidence provenance and sufficiency gates enforced before generation, verification dependency degradation, immutable decision records with an append-only lifecycle log whose current state is derived rather than stored, a ratified contract amendment process, fail-open telemetry with explicit population provenance, sample-size-honest metrics, and a deterministic offline evaluation harness driven by a human-authored synthetic golden set. Every contract rule maps to at least one test, enforced at build time. The artifact runs with no model provider, no network access, no database, and no credentials. Every example, fixture, evidence passage, actor, and evaluation case is synthetic and newly authored. Scope. Its design is informed by lessons from a working production enterprise AI platform, while the implementation, examples, contracts, and evaluation data included here are publication-safe and independently reproducible. The repository excludes the commercial platform, proprietary integrations, production prompts, operational data, customer information, and internal deployment architecture. Evidence classes. Results published with this artifact are labelled "Reference implementation synthetic baseline". They describe this implementation's conformance to its own contract. They are not a measurement of any production system and are not numerically comparable to the field observations reported in the accompanying paper.

Zenodo mints the archive DOI when the GitHub release is ingested; the DOI is therefore not self-declared in this file. It is written back into CITATION.cff and the paper after ingestion. See docs/release-process.md.