Hi HN,

I’m excited to share [Lambda³ Jump Event Detector](https://github.com/miosync-masa/bayesian-event-detector), a minimal yet interpretable open-source library for detecting jump events in time series, using a Bayesian approach inspired by Lambda³ theory.

*Key Features:* - Decomposes time series into smooth trends and discrete jump events, with full posterior explainability. - Directional event detection (positive/negative), not just “changepoint” labeling. - Transparent, factorized output (no black-box predictions): each parameter is interpretable (jump magnitude, uncertainty, trend, event probability, etc). - Super lightweight – core model is just a few lines of PyMC code. - MIT License, ready for customization and extension.

*Why?* Most time-series models treat discontinuities as “noise.” In science, engineering, finance, or biology, those “jumps” are often the real story. This model aims to separate, explain, and quantify those events, with full Bayesian uncertainty.

*Demo (Colab, plots, and results):* → [GitHub repo](https://github.com/miosync-masa/bayesian-event-detector) → [Preprint and theory background (Zenodo)](https://zenodo.org/records/15672314)

> *Background:* > I’m releasing this because, honestly, in Japan hardly anyone noticed or cared about this work (lol) – so I’d love to get feedback from the international/HN community, especially anyone interested in interpretable AI, anomaly/event detection, or Bayesian modeling!

Thanks HN!

— P.S. There are sample experiments, code, and discussion of limitations (no overclaiming). The code is MIT-licensed for both academic and practical use.