This page explains what we publish, what we refuse to publish, and how you can check either. It exists because a text-tools site can very easily present guesswork as fact, and we would rather be explicit about where the line sits.
How the transformations are produced
Every style is a deterministic mapping from an input character to a Unicode code point. Those mappings live in UnicodeStyleService on the server and in the browser engine, and the two are tested against each other: the same 164 styles are run over uppercase, lowercase, digits, spaces, punctuation, accented characters and non-Latin script, and the outputs must match exactly. Nothing is generated by a model or inferred from appearance.
How character coverage is measured
The coverage figures on generator pages are computed at render time by running each style over the complete A–Z, a–z and 0–9 ranges and counting how many characters actually change. They are not estimates and they are not maintained by hand, which means they stay correct if a mapping is ever edited. The Unicode block cited beside each style can be checked against the published code charts.
How platform compatibility is handled
We do not publish platform compatibility unless someone has tested it. A publishable result records the platform, the exact field or surface, the outcome, the device and operating system, the app or browser version where available, who ran the test, and the date. Results older than six months are flagged for re-testing, because app behaviour changes without notice.
Where no test has been recorded, the page says that plainly rather than showing an empty grid or a plausible-looking guess. A compatibility table that has not been earned is worse than no table at all, because it looks authoritative.
What this site does not claim
Unicode defines characters and their properties. It does not guarantee identical glyph design across fonts, acceptance by any particular application, or how a screen reader will pronounce a character. Our guides separate encoded-character facts, which we can demonstrate, from platform behaviour, which requires testing. Where a guide touches on platform behaviour without a recorded test, it describes the mechanism rather than asserting an outcome.
Unicode reference data
Character names, categories and block assignments should come from the Unicode Character Database rather than being inferred. Production deployments can import it so reference pages cite an authoritative dataset and record which Unicode version they used.
Corrections
If something here is wrong, send the affected URL and a short description through the contact page. Corrections are applied to the page and its review date is updated. We would rather fix an error than defend it.