shaunpaull/hypermorphic-sdrg-causal-metric-flow

This is a research prototype in Python that explores a different way for a system to remember and predict things over time. Most prediction systems decide how similar two data points are using one fixed rule, similar to measuring straight line distance on a map that never changes. This project instead lets that rule of similarity itself change over time, based on what has recently happened, so the system remembers by adjusting how it measures things rather than by storing a fixed list of facts. The core mechanism keeps several different ways of measuring similarity at once, called metric charts. One chart pays attention mainly to position, another to speed, another to acceleration, another to how sharply something is turning, and so on. At each moment, the system blends these charts together based on how well each one has been predicting correctly recently, giving more weight to whichever way of measuring has been working best lately. This blended measurement is then used to find past situations that look similar to the current one, and to predict what happens next based on those similar past moments. The repository includes a runnable demo that generates a synthetic scenario where the underlying pattern secretly switches between different modes over time. It then compares this adaptive approach against several fixed, unchanging ways of measuring similarity, printing out accuracy scores and saving charts showing the results, including how much better the adaptive approach performed. The demo can be run directly with Python, given the numpy and matplotlib libraries, or pasted into Google Colab as a single cell. This project is explicitly a research demonstration rather than a production tool or a claim of state-of-the-art results. The README goes into heavy mathematical detail about how the system works, and does not mention a license, so the terms under which the code can be reused are not stated.