Seed text

The seed text is the first thing your coin ever reads. 20 to 200 KB of UTF-8 text. 90% of it is training data; the last 10% is the fixed validation split, so the validation loss stays comparable for the coin's whole life.

Filters applied at upload

• Lines with a slur are removed (leetspeak, spacing and look-alike letters are normalised first).
• URLs, domains, wallet addresses (base58, 32 to 44 characters), @handles and scam phrases (airdrop, claim now, connect wallet, dm me, send sol, presale and similar) are stripped.
• A corpus about one real living person is refused. Rules: (a) one well-known living person named 5 or more times per 20 KB (list: Wikidata humans with 45 or more Wikipedia sitelinks and no date of death, CC0), (b) one first and last name in more than 1.5% of lines, (c) one @handle more than 10 times. The refusal names the rule, never the person.

Public-domain packs

Aesop's Fables (Project Gutenberg #21), Alice's Adventures in Wonderland (#11), Grimm's Fairy Tales (#2591). Each passes the same filters and is capped under 200 KB.