This release forms part of the Computational Observerhood Labs of Mirror Programme, Volume I: Observerhood. Lab I tests whether explicit self-model reliability improves viability under self-relevant perturbation. A controlled grid-world compares four architectures: a predictor-only agent, a memory agent, a self-model agent without reliability tracking and a Mirror reliability agent. Across 6,000 deterministic episodes, the Mirror agent does not dominate all conditions. Its strongest advantage appears under false self-location, where mean viability rises from 139.2 for the strongest non-Mirror baseline to 223.6, and survival rises from 7.7 percent to 63.0 percent. Under generic sensor degradation, the same mechanism becomes maladaptive because it treats unstructured noise as self-model failure. The release includes a standalone paper and a reproducibility package containing the Python implementation, fixed-seed outputs, summary data, figures, requirements, citation metadata and licences.