Paste the charter. Get the one it should have had.
Paste what you would put in a bot profile. You get back what is already there, and the lines to add.
Nothing you paste leaves this browser.
The interface is translated. The profile has to be in English, the language the rules read.
Already there
Missing
Against the corpus
Questions
Where does my text go?
Nowhere. It stays in the tab. The page makes no requests except for its own files, sets no cookies and runs no analytics. Two local storage keys remember your language and your theme. It works with the network off.
What is the corpus?
1102 English charters from 40 public GitHub repositories, every source pinned to a commit sha. It covers what the audit found and nothing beyond that.
Which rules are counted?
3 of the 8 rules cleared the publication gate against blind hand labels and are counted across the corpus. The rest are reported from a sample of 60 on the method page. The secret-shape rule is left out on purpose: its count at this corpus carries no information.
Do I get a score?
There is no score. You get a list of what is there and lines for what is not.
Which languages does it read?
English. The interface is translated, the eight detectors read English only. Text in another language gets a notice under the box and the results are shown anyway.
Does this predict how the bot behaves?
It reads the text. Whether the bot obeys it is a separate question: a charter with an approval line can still be run without one.
What if my charter is shorter than 40 words?
It still runs. The word count flags it, because the audit corpus starts at 40 words.
What does it miss?
Unusual wording. The rules are regular expressions over normalized English text, so an approval gate phrased in a way the corpus never used goes unseen.
Who made this?
One person, independently. It reads text people published themselves and never touches x.ai.
This reads the text. Whether the bot obeys it is a separate question.