The RLM core, compiled straight from Rust to WASM. Deterministic primitives run on your browser's CPU, and this demo shows on real hardware that they match the server (native) byte for byte.
Change the input and seed, then hit "Compute". Same input and same seed gives the same value, no matter how many times you press it or which device you use.
Build the RLM Hot Shelf (Treap, claim 7) 100 times and check whether every fingerprint matches.
An ordinary probabilistic LLM would scatter into 100 different outputs here. SlimeTree-RLM comes out 100/100 identical.
The values below have been verified to be completely identical across verify (native Rust), wasm32-wasip1 (wasmtime), and this page (in-browser WASM). Recompute and show the browser-side values.
This is the "Basic (RLM core alone)" build. β Integrated build (with S1βS9 normalization):
a demo that folds different phrasings, word orders, full-/half-width forms, and English vs. Japanese into "same meaning β same answer".
Standalone tree: AI/SlimeTree/slimetree-wasm-8802/ (impl_v5 is referenced only, unmodified / the existing Python sandbox is also preserved).