This release forms part of the Computational Observerhood Labs of Mirror Programme, Volume I: Observerhood. Lab VI extends the Mirror Observerhood Labs from simple non-linguistic agents to language-like neuro-symbolic agents. The experiment does not evaluate a production language model. Instead, it simulates a stochastic semantic front-end that emits candidate propositions about goals, roles, memory, tools and self-state, then compares four architectures across 38,400 deterministic episodes. The central question is whether a persistent symbolic layer with reliability-weighted commit gates and diagnostic repair can protect action-relevant self-state and tool commitments under semantic corruption. Four agents are compared: a stateless semantic agent, a memory-only agent, a persistent self-model agent and a Mirror neuro-symbolic agent with channel reliabilities, symbolic commit gates and diagnostic repair. The result is deliberately mixed. The Mirror neuro-symbolic agent improves task success under several perturbations, especially false self-state, tool unreliability, memory injection, mixed stress and high repair cost. Net viability, however, improves only under tool unreliability and mixed stress. In other conditions, diagnostic repair and gating costs outweigh the avoided semantic error. The central finding is conservative: a language-like agent should not treat fluent semantic output as durable state by default. Candidate propositions should pass through reliability-weighted symbolic commitment, contradiction checks and cost-sensitive repair. But this architecture is not automatically beneficial. It becomes viability-positive only where semantic corruption threatens action and where repair cost is justified by avoided loss. The release includes a standalone paper and a reproducibility package containing the Python implementation, fixed-seed outputs, raw per-episode results, summary data, Mirror viability advantage data, success advantage data, figures, requirements, citation metadata and licences.