The lesson
Introduce a benchmark by naming the gap first, then show its construction, sample tasks and scoring rules before sending readers to the full paper.
The article opens with the title, a one-line claim about a reasoning benchmark, metadata, contents and related research links. A “TL;DR” summarizes FormulaOne as “220 novel dynamic programming problems over graphs” in three tiers, then points to the full paper. The body explains why doubleAI built the benchmark, shows a tier table for Shallow, Deeper and Deepest, gives example problems from each tier, then explains Courcelle’s theorem, tree decompositions and the evaluation checks: Correctness, Consistency and Efficiency.
Don't copy FormulaOne, the graph-theory framing or doubleAI's benchmark claims. Keep the structure of gap, benchmark design, sample tasks, theory explanation and evaluation rules.
Tell us what you want to make. We’ll talk through how to put these ideas to work for your brand.