This is not a good kata idea, period.

Translating one language to another is itself already an nigh un-automatable job, not to mention this is translating from a phonetic language to a written language. What real people do with these is to manually decipher each word from the particles that vaguely pronounces into them by the surrounding phrase, not doing it programmatically. In fact, doing it programmatically is impossible: there are many words in English with the same pronounciation, which you cannot hope to be distinguished by phonetic representation.

And you require user solution to translate almost half of a very large corpus to an exact result. What do you expect us to be, omnipotent gods? Not even the best neural networks (which are the best algorithms btw) can perform such translations with 100% accuracy. This is beyond unreasonable. I can also guarateed that the coded rules will easily break or translate incorrectly given words unseen before. (In fact, you already did. The last test want INDEPENDANTISTS when the correct spelling is INDEPENDENTISTS.)

In short, the task being asked in the kata is impossible. So to pass the tests one will have to cheat somehow, somewhere.

Also, I like how even the author solution says Even Finkel cheated in his alphabet as half of the sounds are not existing!. If the original decipherer needs to manually add tons of rules to cheat the relationship, there is no hope to making it a kata. Nobody shouldn't be forced to do the same kind of hacking.

(Also, QAnon in the last test? WTF? Is the kata author drunk or something? Nobody should expect these kind of things to be translated correctly.)

The kata does not provide enough rules on the task given, and the only link given with the rules (the Wikipedia article only lists the cuneiform signs) is to a website where the media explaining the rules is behind cookie wall (Please allow marketing cookies to view media). Please provide a different source.