Werr

No. Jev answers from a hosted model while werr synthesizes decisions locally from geometric wave and recursion logic. Treat them as different architectures sharing only the typed-answer shape.

It targets zero VRAM CPU execution with no weight tensors to load. Check the live browser simulator and the benchmark directory to see what that tradeoff means for your workload.

You can try it with nothing installed. The interactive decision simulator runs client-side in the browser, and Python 3.10 or newer is the stated requirement for running the code itself. Accuracy claims are not documented in the README; the benchmark and audit reports under benchmarks/ and docs/ are where the numbers live, and JevBench results are its own published runs rather than an official leaderboard.

The repository uses a dual BSL and MIT license file. Read the license file before embedding it in commercial software.

No. It reports zero bytes of VRAM and derives decisions from geometric waves, Euler thresholds, and recursive subdivision rather than 4B-parameter tensors. It runs on Python 3.10 or newer, and the badge shows version 0.5.1.

The README defaults to English with a Turkish translation in README_TR.md, and the GitHub Pages decision simulator supports both English and Turkish with light and dark modes.

A Hugging Face dataset at pCwOrM/werr_open_decisions, a live benchmark arena on GitHub Pages covering WebMCP, JevBench, Gym, Arena, and Tau-Bench, plus the accompanying arXiv paper 2609.25498 with a Zenodo archive.