Editorial policy

One question, answered first

Each page answers one implementation question in its opening paragraph, then shows a complete original example and walks through the cases that usually break. Every page ends with what it does not cover and its sources.

Source, result, observation, inference

Every page keeps four kinds of statement apart: what a source documents, what a calculation produces from stated inputs, what we observed in a named browser and configuration, and what we infer. Planned work stays visibly planned. Anything we did not run stays “not tested”.

Measurements name their conditions

A performance statement identifies the metric, browser version, operating system, device class and relevant display conditions. A single trace is not a universal speed claim. Emulated touch, an emulated motion preference and accessibility-tree inspection are labelled separately from physical devices, operating-system settings and real screen-reader output. Passing a test page or a scanner does not establish accessibility compliance, and we make no compliance claims.

Original work only

Explanations, examples and test pages are written new for this site. We do not reuse prose, code, demos, images or data from earlier sites at this domain; owning a domain is not a licence to old content.

AI assistance and responsibility

Pages are drafted with AI assistance. The publisher is responsible for every page: claims are checked against their sources and test pages before publication, and errors are corrected in public as described in corrections.

Review

Independent technical review by a named frontend engineer is being added. A reviewed page shows the reviewer’s name, credential and review date. Reviewers must be independent of any sponsor on the page.