https://badecision.app.xerios.dev

This started off mostly as me experimenting on training small decision models to see what kind of challenges are involved, how can I control and refine the output, which methods there are available and how far can I push them. Despite this ending up as a pure parody, this isn't entirely a nonsensical project. I've spent quite some time on this, so it isn't a one-off prompt.

Of course I doubt this project would have much real-life use due to its chaotic nature, but there's some fun in doing the complete opposite of what everyone else is doing. In the end, it's a bit of a mentally-unstable decision model that can still reason to some degree.

For me it's the journey and the parody website that mattered, not the final results. That said, everything related to the validity, accuracy and safety of building the model is mostly dependent on Claude Opus 5.5 (high) reasoning, planning and implementation (with some of my own guidance).

As for the published package: it is zero-dependency, client-side and wasm. As safe as it can be.