The lesson
Open a benchmark report with one plain question and a chart the reader can re-cut by tokens, cost, time or output before any prose.
A breadcrumb 'BENCHMARKS / CYBER', a 'PROPRIETARY' tag and 'UPDATED 11/11/2026' sit over the title 'CyberOffense' and the question 'Can AI autonomously compromise realistic enterprise networks?'. A 'NOVA + LIGHTHOUSE + WATERSHED: SCORE VS. CUMULATIVE TOKENS' chart has TOKENS, COST, TIME, OUTPUT and LINEAR, LOG toggles plus COPY IMAGE, followed by a 'RESULTS ON EACH RANGE' table showing averages of 3 runs with their spread, then one chart per range. 'Key Takeaways' opens each point with a bolded sentence such as 'The cloud range is the hardest of the three.', and 'Background', 'Key concepts' and a 'Methodology' diagram of the attacker machine, cyber range and scoring follow.
Don't copy Vals AI's model scores, range names or cyber framing. Keep the order of question, interactive headline chart, per-segment results table with run spread, bolded takeaways, then plain-language background and a methodology diagram.
Tell us what you want to make. We’ll talk through how to put these ideas to work for your brand.