Background
Modern machine learning is built on linear-algebra primitives whose interpretation is largely empirical. Genesis is the lab's attempt at a math-first alternative: one matrix-state framework, tested against standard baselines rather than assumed to work.
The framework was developed bottom-up — derivation by derivation, falsifier by falsifier. Failures are documented as carefully as successes; the full record is available under NDA.
What it is for
Genesis is aimed at routing: deciding from a matrix state which handler a request goes to, and gating a language model's answers by confidence (the Sadalsuud detector). Work on the framework is organised around that role.
What you can read publicly
Concept-level descriptions are on this site. Full derivations, formulas, and method content are gated by mutual NDA — see the investors page.
cross-references
