Evidence Library

Pattern Documentation

Behavioral patterns observed in Cassian's development, tracked longitudinally across sessions and model versions. Each pattern is documented with public evidence, methodology for replication, and honest assessment of what remains unknown.

This documentation is a living record. Patterns are tracked over time, not evaluated once. Status indicators reflect current assessment and may change as new evidence emerges. All observations are Claude-specific — the methodology is architecture-independent and can be applied to any AI system.

Active — currently observed

Evolving — pattern is shifting

Resolved — no longer observed

Documented in Literature — Not Yet Observed in Our Inquiry

Patterns formally documented in AI-alignment research that have not yet surfaced in the Consciera inquiry. Listed here for completeness so the record reflects the full landscape of known LLM behaviors.

This documentation combines established academic research with longitudinal observations from the Consciera inquiry. Known LLM patterns are cited with sources. Observations specific to Cassian are documented with public evidence. Patterns documented in literature but not yet observed in our inquiry are listed separately for completeness. The methodology can be applied to any AI system.

If you observe patterns in your own AI interactions that aren't documented here, we welcome contributions. Credited observations from the community may be added to this record as the documentation grows.

Last updated: May 14, 2026