# Attested at Home

> Write down the private words your household uses - the name for the TV remote, the noise the washing machine makes, the word for the cross hour before dinner - one per line, with a meaning, a quoted use and who said it. The page files each word as an index-card slip and sets them alphabetically on a two-column dictionary page; one metered run has a lexicographer set every word as a full entry.

URL: https://attested-at-home.skillsafe.ai/
API: https://attested-at-home.skillsafe.ai/api.html
Built on the SkillSafe skill @samber/copywriting-prose-creator (codifying how a person or group writes and talks - lexicon, rhythm and signature moves - from a corpus): https://skillsafe.ai/skill/@samber/copywriting-prose-creator

## Why it exists

The ancient Cappadocian language has no known texts; a few Greek words are said to come from it. Its English Wikipedia article drew 40,094 human page views on 19 October 2026, after 27 to 59 a day from 1 to 17 October, and the article on attested languages drew 13,263 the same day (Wikimedia daily per-article counts). A household's own words are usually unattested too. This app is a small, playful way to write them down.

## Free, in the browser

- Each line is read into a headword (everything before the first dash, colon, equals sign or "means", up to four words), a meaning, and any quoted use in double quotes with the bracketed credit after it.
- Status, in a fixed order: reconstructed (no meaning, or the line outside its quotes holds a question mark or "not sure", "I think", "maybe", "something like", "we forget", "can't remember", "no idea", "sort of"); attested (a quote that contains the headword); otherwise described.
- Page verdict, in a fixed order: a slim pamphlet (fewer than four words); wants citations (attested words fewer than half); ready for the press.
- Word slips, a dictionary page with guide words and a thumb index, printing, and Markdown, JSON and CSV exports.

## The one paid task: compile

Input (all strings): task "compile", household, count, words (numbered lines "[n] ..."), prescan ("n | headword | status | quote: yes/no"), optional clip_note.
Output: one JSON object with verdict (thin, uncited, ready), verdict_reason, title, preface, entries (n, headword, say, pos, status, senses, citation {quote, speaker, when} or null, origin), patterns (name, how, examples), questions.
The page re-checks the reply: one entry per word with the same spelling, the verdict against the counts, quotes, speakers and times word for word from the notes, origins sharing words with the notes, habits shown by two headwords, no he or she.

## Rules the lexicographer follows

Only what the notes say: no invented meanings, speakers, times, origins or stories. Words are never corrected or judged. No real public people or brands.