TimeCopilot/impermanent

Impermanent attempts to create a 'Chatbot Arena' equivalent for time-series foundation models (TSFMs). While the concept of a live benchmark that measures drift (temporal generalization) is timely and valuable given the rise of models like Amazon's Chronos and Google's TimesFM, the project currently lacks the necessary community traction or data gravity to serve as a moat. With only 7 stars and 1 fork after a month, it remains in the prototype phase. Its defensibility is low because the evaluation harnesses for these models are already being standardized by the labs themselves (e.g., Nixtla's benchmarks or the 'Chronos' evaluation scripts). The primary value of such a project lies in its status as a neutral third-party authority; without massive adoption, it is easily displaced by established time-series ecosystems like Nixtla or specialized ML observability platforms (e.g., Arize, WhyLabs) that could implement 'live drift' benchmarking as a feature. The tech stack is a standard wrapper around existing TSFM libraries and statistical packages.