FRAC presale: check the current server quote, availability and allocation terms — Buy Now

📈 Proof-of-Training Ledger

The signed learning history of an AI model — written inside consensus, on every block.

FractalAI is the only Layer-1 whose consensus (Proof-of-Useful-Work) writes a real training step into every block header — model id, loss before/after, improvement — Dilithium-signed by the validator. The chain becomes an append-only, tamper-evident ledger of a model’s loss curve, re-verifiable by anyone against the public RPC.

Honest scope: proves these loss/accuracy numbers were signed into these blocks in this order — NOT the dataset, NOT absence of bias, NOT that a production model equals the registered one. Models on the node today are small (FCNN), not LLMs. Frente K · Motor de Jugadas.