The lesson
When you publish a benchmark you won, show the tasks, the rubric weights and your own bias openly, so readers can check the result instead of taking it on trust.
Under 'Real-World Performance, Not Lab Scores' the page states the three scores up front: Strawberry 99.2 out of 100, Comet 90.8, Atlas 73.3. 'What We Tested' lists 12 benchmarks in three tiers, each a concrete job like sourcing 150+ leads from LinkedIn Sales Navigator and deduplicating against a live CRM. 'How We Scored Them' gives the point weight of each of five dimensions, from Data Accuracy (35 points) to Speed & Efficiency (5 points), followed by an 'Evaluation Approach' note on LLM-as-a-Judge scoring. 'Addressing Bias: Why These Findings Matter' then admits the brand designed the test, and 'The Results in Detail' explains each browser's score in plain terms, such as Atlas pulling C++ repos instead of TypeScript.
Don't copy Strawberry's tasks, scores or competitor claims. Keep the order of headline result, task list, weighted rubric, method caveat, bias section and per-competitor explanation.
Tell us what you want to make. We’ll talk through how to put these ideas to work for your brand.