Making-of, training notes, and real results from the Syvorai AI Lab models.
The judge bake-off against the human gold, the hybrid distillation of v6, the v7 to v12 experiments, the nine-challenger campaign, and the champion v10+swa ensemble published on Hugging Face under Apache 2.0.
The ruler was wrong: a paired benchmark against the zero-shot backbone, blind adjudication of 175 disagreements, a gold set of 1.000 tickets, and the v3 → v4 → v5 trilogy up to the root-cause fix — re-labeling the entire train set with a strong judge.
The results of decisor v2 and v3: gold set accuracy, P/R/F1 of the 17 categories, calibration via temperature scaling, three RLCD rounds with a negative verdict, and the audit that revealed the evaluation ruler was wrong.
The beginning of the Decisum project: why deciding is different from conversing, the System 1 + System 2 architecture, the 447 thousand examples from three sources, the AMD ROCm infrastructure, and the tokenization bug that nearly buried the first training run.