A ~20,018-row chain-of-thought finance-reasoning dataset for supervised fine-tuning — the exact
training set behind the companion model (Mixtral-8x7B-Instruct LoRA, lead entry E24). Built as the
data-recipe submission to the AutoScientist Challenge
(Finance, Part 1, Adaption Labs, 2026) and released under CC-BY-4.0.
It is a 4,018-row curated seed (Adaptive Data quality 9.3/10, grade A) expanded to ~20K with… See the full description on the dataset page:
https://huggingface.co/datasets/tejadhith/asc-finance-reasoning-20k.
