130 instruction→JSON pairs that teach a small model to break one overwhelming task into tiny, timed, categorised ADHD-friendly steps, in the exact schema used by Unstuck.
Each pair was distilled from a strong serverless model (Qwen/Qwen3-30B-A3B-Instruct-2507 on Nebius Token Factory) prompted with Unstuck's real breakdown prompt across 44 tasks × 3 granularities (chunky / regular / tiny), then filtered through… See the full description on the dataset page:
https://huggingface.co/datasets/art87able/unstuck-sft-breakdowns.
