Corpora
The playground provides three English corpus choices. Each choice is processed with the selected alphabet before analysis.
Monkeyracer
Section titled “Monkeyracer”Monkeyracer is a deduplicated collection of quotes from TypeRacer and Monkeytype. It represents the quote text used by those typing sites, rather than a word list from Monkeytype.
Reddit is based on ConvoKit’s Reddit corpus, which sampled conversations from 100 highly active subreddits in September 2018. Layouts Wiki removes markup and URLs, normalizes whitespace, converts curly single and double quotes to their ASCII forms, decodes a limited set of HTML entities, shortens long repeats, then takes a fixed-seed whole-line 5 MB sample from the 55,629,660-byte cleaned source.
Wordfreq
Section titled “Wordfreq”wordfreq 3.1.1, maintained by Robyn Speer, combines English data from Wikipedia, subtitles, news, books, the web, Twitter, and Reddit. Layouts Wiki uses its Small English vocabulary, which keeps entries at least 1 per million in the upstream frequency list. Entries retain their weights and are analyzed as independent words: no spaces or transitions are inserted between them, and a word is discarded when any grapheme is outside the selected alphabet. Numeric aggregate templates are excluded. The derived data is available under CC BY-SA 4.0; see the upstream source and attribution.
Processing
Section titled “Processing”Every corpus converts curly single and double quotes to their ASCII forms before applying the selected alphabet. Monkeyracer is normalized statically into one complete source. The browser downloads that source once, then applies the selected alphabet locally by retaining a complete quote only when every character belongs to it. Changing the corpus partition or alphabet reuses the downloaded source. Build-time populations apply the same whole-quote policy. Wordfreq applies a whole-entry rule to each independently weighted entry. Reddit instead cuts its prose into supported sequences at characters outside the alphabet and drops resulting fragments shorter than 32 graphemes. Case and spaces remain unchanged. Numeric aggregate templates are excluded from Wordfreq even for the ANSI alphabet. The alphabet choices therefore change which text is retained and which layouts have eligible statistics.