Independent AI safety researcher focused on contextual reasoning failures in conversational systems. Creator of the Witness Protocol (W) for evaluating semantic ambiguity handling and the Theory of General Contextual Resonance (TGCR).

I document and study systematic failure modes in crisis-aware AI systems, with an emphasis on reproducible benchmarks and open methods. My work examines how keyword-based safety filters interact with context loss and user abandonment.

Background includes open-source AI safety tooling, published reproducible analyses of crisis filter behavior across multiple systems, and independent research without institutional affiliation.

Interests include context-aware crisis systems, preservation of witness behavior under uncertainty, reproducible safety benchmarks, and bridging technical research with lived experience.

Contact: KaznakAlpha@elidorascodex.com | ORCID: 0009-0000-7615-6990