This report presents the full technical and legal analysis accompanying the OE-EV-2026-01 dataset release (9,430 trials), applying Lean Six Sigma and reliability-engineering process-control metrics (DPMO, Propositional Error Rate, Mean Time Between Failures, Severity-Weighted Defect Rate) to fabrication behaviour in four commercial and open-weight language model architectures under controlled epistemic conditions. Across 9,114 scoreable propositional claims, the study finds a population-level Propositional Error Rate of 22.61 per cent, with a 24-fold range in fabrication rate between architectures under identical, grounded (FEASIBLE) conditions and identical adversarial pressure (DIALECTIC protocol). The report documents this cross-architecture variance as evidence bearing on the Reasonable Alternative Design standard under product liability doctrine, and separately reports a failure mode ("the Beta Inversion") in which adding an Input Sanitisation Node to the best-performing architecture increased its fabrication rate under adversarial conditions, a finding the report addresses directly rather than omitting. The report includes a full methodology appendix, tamper-evident SHA-256 chain-of-custody hashes for all artefacts, and a disclosure statement covering the author's commercial interests and intellectual property holdings. The written analysis is licensed under CC BY-NC 4.0; the accompanying evaluation scripts are dual-licensed under AGPL-3.0 or a separate commercial licence. This is an independent technical report, not externally peer-reviewed. It does not constitute legal advice and does not predict the outcome of any litigation; the legal frameworks discussed are theoretical analysis, not jurisdiction-specific counsel.