Methods

Scope of Observation

This archive studies interaction-layer dynamics of large language models under explicit constraints. Claims are limited to observable behavior in turn-based interaction and do not extend to internal model representations, training processes, or semantic correctness unless explicitly stated.

Constraint-First Methodology

All experiments and theories in this archive are generated under explicit constraints, including but not limited to token limits, throughput envelopes, fixed prompts, replay conditions, and termination criteria. Constraints are treated as structural features rather than noise to be optimized away.

Invariants and Derivation

Observations are elevated to invariants only when they remain stable across perturbations of prompt phrasing, sampling randomness, and interaction ordering. Examples and narratives are used solely as derivation traces and are not treated as claims.

What Counts as Evidence

Evidence in this archive is limited to:

• Interaction traces
• Failure and collapse depth
• Stability across repeated runs
• Behavior under replay versus looped interaction
• Persistence under perturbation

Explicit Exclusions

This archive does not make claims about:

• Optimal model usage
• Organizational best practices
• Ethical alignment or safety guarantees
• Commercial deployment outcomes
• Psychological diagnosis or intent attribution

Relation Between Experiments and Theory

Experiments establish observable structure within constrained interaction space. Theory abstracts invariant patterns from those structures. No theory is introduced without experimental grounding or external observational support.