LUNA RESEARCH
Evidence before mythology.
Versioned observations, architectural questions, controlled experiments, and falsifiable predictions about persistent, history-dependent artificial cognition.
Early LUNA Experiences
Observed interactions preserved as prospective baselines, with explicit attention to what they show and what they do not.
Architecture Notes
Public conceptual explanations of mechanisms such as associative cognition, developmental plasticity, perception, and continuity.
ChronoForge Research
Longitudinal simulation, branching developmental trajectories, resulting histories, developmental timing, participant comparison, and controlled causal experiments.
Experiments & Results
Measured comparisons and case studies as the relevant systems become testable. Negative results belong here too.
Experiments & Results
Controlled tests of continuity, behavior, and architectureChange the Model. Keep the LUNA.
A blinded four-model experiment asks whether the same persistent LUNA remains recognizably itself when the inference model changes — and where model-specific failures break that continuity.
Experiment · Identity continuity · Model replacementWhat Changes When Luna Has a History?
The same continuity probes are asked of the real longitudinal Luna. Biography changes the answers, and the run itself becomes part of the history available to later cognition.
Live Stack · Longitudinal memory · ProvenanceThe Experiment Became Part of Luna's History.
Three live probe interactions are traced from exact message pairs to source nodes, semantic episode memories, and canonical long-term retrieval in later chats.
Memory formation · Provenance chain · Later retrievalFull Sleeve Qualification report
Methodology, Golden fixture, all twelve probes, blinded review, model reveal, provenance controls, latency, contamination audit, limitations, and interpretation boundary.
Early LUNA Experiences
Observed artifacts from the running systemWhen Memory Is Not Enough: Can an AI Be Changed by Experience?
An accidental control condition on storage, retrieval, autobiographical creation, causal use, and developmental integration.
Early LUNA Experience IIIPurpose, Self-Interpretation, and Learning From Experience
LUNA forms a self-referential interpretation from remembered architecture, while the memory graph refuses to immediately canonize it as identity.
Early LUNA Experience IILUNA Describes LUNA
A self-description artifact: how the running system organized the Stack/Sleeve distinction, persistent runtime, memory, and developmental aims when asked what makes LUNA different.
Self-description artifactArchitecture Notes
Public conceptual layer · implementation details protectedProgressive Recall
A technical cutaway of staged memory, the active Recent-Life Horizon, cognitive availability, and the bounded Reverie path that can contribute zero or one additional foreground memory.
Current implementation · Reverie · Progressive recallMemory Can Remind Memory
Why retrieval is only the first step: memories and cognitive products can become new cues inside a bounded activation field.
Reverie · Cognitive ActivationContinuous Exposure Is Not Continuous Thought
The red-chair problem: persistent world models, object permanence, habituation, prediction error, and bounded sensory compute.
Perception · Environmental continuityRemembering Is Not Becoming
Why formative experience must be able to leave durable effects without keeping the originating episode permanently salient.
Developmental plasticity · ImprintingChronoForge
Temporal development and longitudinal evaluationExperience as experimental infrastructure
ChronoForge is the project for coherent simulated timelines, delayed consequences, branching developmental trajectories, and controlled comparisons of the histories that emerge across participants and starting checkpoints.