Brand · voice
Ten lines we say, ten we never do
A voice guide is only as good as its enforcement. This one is a two-column table with a check: the right-hand column is filled with the same phrases the tone scan searches this repository's prose for, and the scan's live numbers are printed underneath.
We say
We never say
2.1 KB, zero dependencies.
Seamless, effortless integration.
A number is checkable in the file; an adjective is a promise about the reader's experience, which we have not met.
Reduced motion is designed, not disabled.
Accessible by default.
The second is a badge; the first is a decision somebody can inspect in the stylesheet.
This one is 41 KB heavier; here is why we kept it.
Optimised for performance.
Every component is somebody's trade-off. Name the trade, especially when it went the expensive way.
Not measured on this build.
Best-in-class performance.
The honest gap is more useful than a grade, and it survives being checked.
Three of the 74 prompts were dropped for reproducing poorly.
A hand-curated collection.
Selection with a rejection rate is curation; selection with no numbers is a synonym for 'we chose some'.
The demo scenes are 15,653 lines across 17 modules.
Lightweight.
Lightweight compared to what? A count is comparable; a word is a mood.
Try the keyboard walk: unplug the mouse and finish a task.
Fully accessible.
Hand over the test instead of the verdict — the reader can run it in a minute.
This page does not claim a Lighthouse score, and says why.
Blazing fast.
Also on the banned list below, because 'blazing' is the adverb a dashboard uses when it did not measure anything.
Copied 3,910 times, and 128 people kept it in a stack.
Trusted by thousands of developers.
Real counters we can actually read, versus a population we cannot.
We did not build accounts; a demo would have meant fake data.
Coming soon.
Naming the boundary is a decision; 'coming soon' is a decision deferred onto the reader.
The banned list, in full
18 terms, imported on this page from the same module the scanner reads — so this list cannot drift into a shorter, friendlier version:
Two things keep the list from being theatre. A hit within 90 characters of a negation is treated as intentional and cleared — which is how this page explains why we ban “magical” without flagging itself. And the scan runs over the site's own source at build time, so the numbers below are facts about this repository rather than a policy statement:
tone scan, run against this source tree at build time: files scanned: 138 banned-term hits: 7 cleared as negations: 7 live hits: 0
The numbers are read from the scan at build time, and the export harness asserts the live-hit count is zero — so a page that adds a banned word fails in the same commit that adds it, including this one.
Related: the logo system and the audit register.