STOLZ A.I. and the Discipline of Less

Field notes

The system separates route selection, context loading, result reuse, quiet-state observation, and benchmarking into five independently named skills. Its evidence policy requires the baseline and optimized routes to receive the same versioned task, produce equivalent required outcomes, and pass verification before any efficiency delta is accepted. The homepage also describes safe fallback behavior when capabilities are missing, while noting that the system does not replace the model, runtime, or project-specific checks.