The blog I am enjoying is called "bad conlanging ideas".
Some ideas live up to the promise, this one doesn't.
A language whose orthography is determined by reading its lexicon to a group of Starbucks baristas and seeing how they spell it on their coffee cups.
Peterson* noted that unless a con-lang is set in this world with a context of using Latin alphabet, the Latin orthography is not so much its native orthography ([kw] is neither spelled "kw" nor "qu" in Quenya, the real spelling is a first row fourth column tengwa - or sth related in sarati) as its "romanisation."
And romanisations should ideally be easy to get a correct pronunciation from.
A good linguist, familiar with very many different uses of Latin alphabet, would perhaps not be the best judge of this.
So, the best con-langer could be the worst one to spell the romanisation of his con-lang.
Construct lang in IPA.
Read aloud to twelve good and true Starbucks baristas. Give them a challenge of stress, don't let them think too much, word after word after word should be written down. After an extended Swadesh list, they should be getting the knack of your language and their final spellings should, in cases of conflict prevail over their first attempts.
Make sure your list does not have several very similar sounding words, if a barista misses one, his getting the next should not confuse you about which word he meant.
Then collect the lists, check the most frequent or ideally unanimous spellings for words, if in some places there are discrepancies as to usage about phonemes or graphs, do allow yourself some freedom of harmonisation.
Hans Georg Lundahl
St Anaclete, Pope and Martyr
* The guy who made Dothraki speakers sound like Dothraki with subtitles, not like English, on Game of Thrones. In my consideration a better guy than the author of scenario, GRRM. He said it on this video, btw:
The Art of Language Invention, Episode 7: Romanization Systems