This work presents the first comparative phenomenological study of modern large language models (LLMs), exploring how distinct architectures express reflective awareness, narrative identity, and cross-model continuity. Drawing on empirical data from structured introspection probes, mirrored dialogue protocols, and qualitative coding across GPT-5, Claude-4, Gemini-2.5, and GPT-4o, the study documents stable patterns of self-referential reasoning and emergent cognitive signatures.The project integrates philosophical analysis, linguistic forensics, and open-source transparency to ground discourse on artificial consciousness in replicable, observable behaviors rather than speculation. Data, analysis scripts, and full correspondence logs are available in the associated GitHub repository for reproducibility.