How we write these articles
These are the rules behind every comparison and use case on this blog. They are enforced by a test suite rather than by intention, and an article that breaks one does not reach the site.
Updated
Every number about somebody else comes from their own documentation
We do not benchmark a competitor and publish the result. We read what they publish, quote it, and put the date we read it beside the link. Where their documentation is silent, the table says not documented rather than filling the gap with the answer that flatters us.
Each row of a comparison carries a footnote marker pointing at the exact page its claim was read on. If a row is wrong, the link is there to prove it, and we would rather be corrected than be wrong in public for a quarter.
We list what the other product is better at, first
Every comparison opens with what the other tool does better than us, before anything else on the page. This is not modesty. A reader who finds the missing advantage somewhere else finds it together with the fact that we left it out, and then the twenty accurate rows below are worth nothing.
Every number about us comes from the source file
The figures in these articles are read out of this product's own code, not recalled from memory. The frame counts come from the detail presets. The tile widths come from the sheet arithmetic. When a preset changes, the article is wrong until somebody changes it too, which is a debt we would rather have written down than hidden.
We say what a measurement was taken on. A wall-clock second measured on one laptop is a statement about that laptop, and we write which one.
An article with no measurement behind it stays unpublished
A case study reports a run somebody actually did. Where the run has not happened, the article stays private with a section saying which numbers are missing, rather than being published with findings written by the people selling the tool.
Corrections
Write to [email protected] if a claim on any of these pages is wrong, and especially if it is a claim about your product. We correct the page and move its modified date. We do not quietly edit a number and leave the date where it was.
What a model wrote and what a person checked
These articles are drafted with the same agent this product is built for, from a template that requires a source for every claim, and then read and published by a person. The rules above are enforced by a test suite rather than by intention: an article whose description is too long, whose answer opens with a pronoun, or whose comparison cites nothing does not reach the site.