What is epistemic-heat-death-and-the-signal-to-noise-ratio-of-the-global-web?

This repository holds a working paper, not a software application, written by Jarosław Szulc about what happens to the internet as AI generated text, images, audio, and video make up a growing share of new content. The author argues that the web can look more full of content than ever while the actual information value inside it, meaning distinct ideas, grounded claims, and identifiable authors, quietly shrinks. The paper names this pattern epistemic heat death and builds a set of formal tools to describe and measure it, including three governing laws, a mathematical model built from differential equations, and a proposed Signal to Noise Ratio index for measuring how much real information content a document, website, or the web as a whole still holds. Alongside the paper text, the repository includes Python scripts that implement the paper's models so a reader can run them directly and see the same numbers referenced in the text. Separate scripts cover the three laws, the dynamic model and its scenarios, an attempt to fit the model against real web data from Common Crawl, the Signal to Noise Ratio index, a taxonomy of different kinds of information decay, and a small real-data test applied to news coverage of one shared event. Another script regenerates all eleven charts used in the paper from these same models. Running any script only requires standard Python data science packages such as numpy, scipy, and matplotlib. The repository also includes a short, separate technical specification called MVP-0 Content Provenance Spec, which describes a concrete, implementable way for any website to mark its content as coming from a verified, accountable source, using an HTTP header, an HTML tag, and a shared web convention. This project will mainly interest researchers, policy thinkers, and technical writers who care about the long-term health of information on the internet, rather than developers looking for a tool to install. The author is explicit that the paper's central claims are not yet empirically confirmed and invites public comment and challenges to the model. The whole repository, including the paper text, charts, and code, is released under the Creative Commons Attribution 4.0 license, which allows reuse, citation, and building on the work as long as the original author is credited.