03 / RESEARCH
Research queue.
Work we want to test and papers we want to interrogate. A queue records intent; it does not certify a claim.
The research agenda, reconciled.
Our September 26 registry brings 90 experiment concepts into one working plan. The original 56 retain their IDs and ordering; 34 later ideas extend the agenda.
The first planned study remains output-length compression. The wider queue covers model self-observation, group behavior, alignment, internal-state dynamics, memory, representation, audio, robotics, and executable worlds.
Representation Search and Specialized Intellectual Geometry →
Training Pressure, Personality, and Conversational Dynamics →
Build the instruments around the question.
The proposed lab roadmap starts with a reproducible inference and logging setup, then adds internal-state measurements, multi-agent environments, and specialized equipment when a protocol needs them. A roadmap is not a claim that those capabilities are already built.
Cyber studies are confined to harmless, isolated, authorized synthetic environments. Biological and psychological analogies generate hypotheses; they do not establish what a machine is experiencing.
None published here yet. The registry documents questions and proposed tests, not established findings.