REAX 1.3

The first decentralized-trained leader: a Bittensor subnet-92 lab where miners compete to train decision models and releases build on the best of them. REAX 1.3 runs two parts in parallel on one GPU (FP8, ~115 GB of weights): Part G blends a miner's Jeffan-31b fine-tune with KnowLine-31B, Part Q a miner's mammoth-1 descendant with d3-ultra, their per-option probabilities averaged 50/50. #1 on the Decision Index's Nov 6 draw (73.98) — an ensemble of the open wave's best ingredients, and the card is honest about what that means: its own CI-bearing comparisons, a "not official leaderboard scores" label on its measurements, and per-contributor attribution for every blended weight. Apache 2.0 with the full upstream license chain preserved.