Drift and selection in LLM text ecosystems

This project is a theoretical research artifact supporting a scientific paper. With 0 stars and 1 fork, it currently lacks any community momentum or production utility. It addresses the 'model collapse' phenomenon—where AI systems trained on AI-generated data lose variance—by providing a 'exactly solvable' mathematical framework based on n-gram approximations. While scientifically interesting and timely (following the trajectory of high-profile papers like Shumailov et al. on Model Collapse), it is not a software product with a moat. Its value lies in the insights provided to researchers at frontier labs (OpenAI, Anthropic) who are grappling with the exhaustion of human-generated training data. It is highly reproducible once the paper is published. There is no platform risk because this is meta-research rather than a tool for the AI value chain. Its 'moat' is purely the academic depth of the mathematical proof, which is easily absorbed by the wider research community once disseminated.