This is a game based on Wordle, but using words from the Tamil language.
Since Tamil is written using an abugida script (every consonant-vowel/உயிர்மெய் pair gets a distinct symbol), the logic for making and displaying guesses is a bit more complicated than in English. Every "letter" in the game can be either a pure vowel (உயிரெழுத்து, only at the beginning of a word), a consonant-vowel pair (உயிர்மெய்யெழுத்து, inputted as a consonant followed by a vowel), or a pure consonant (மெய்யெழுத்து, inputted as a consonant without a following vowel). Tamil-Grantha consonants are treated as normal Tamil consonants, but the aydham letter (ஆய்த எழுத்து), ஃ, is given special treatment—it behaves like a consonant for the most part, but it cannot combine with a vowel.
When results are shown, each "letter" square can have two different statuses; the colors of the border reflect the correctness of the vowel, and the colors of the center reflect the correctness of the consonant. This leads to some weird edge cases on words which start with a pure vowel, since they may show as both correct and wrong at the same time (i.e., that vowel is present in that square, but there needs to be a consonant before it). For the purposes of displaying results, the virama/pulli/புள்ளி is treated as a vowel diacritic, as is the inherent vowel அ. Pure vowels have their results shown in both the consonant and vowel areas (both the background and border of the square).
Based on the distribution of lengths from ta.wl, a length of four "base"
letters was chosen as optimal. This is actually closer to a length of eight
letters as it would be defined by the English Wordle (up to four vowels and
four consonants), but only certain combinations of letters are valid based on
the phonology of Tamil, so it only feels slightly more difficult than the
English version.
The base word list (ta.wl) is copied from GNU Aspell.
This is piped through UNIX shuf to produce a non-alphabetical version of the
word list (ta-shuf.wl). The frontend filters this list to only words which
match its calculation of length (four "base" letters, either consonants or full
vowels, plus any number of vowel diacritics attached to the consonants), and
indexes into the filtered list using the number of days since the program was
written (February 4, 2022).
Even though there are only four squares displayed on the frontend, results are displayed as eight square emojis; each represents a consonant-vowel pair (the first square in each pair is the consonant, the second is the vowel).
This code is open-source under the GPLv2 license. The word list this program uses comes from GNU Aspell, which is also open-source under the GPLv2 license.
This was inspired by Wordle (obviously) and Shabdle: Wordle in Hindi.