All my early languages aimed at a phonetic representation. Thus I was rather shocked the first time I encountered Dirk Elzinga's wonderful Tepa, which spells things like [tuɣu] as tuku and [yɨška] as yɨyka. But now that I've spent a lot more time staring at Native American languages — including plenty in the Uto-Aztecan family, which seems to be the inspiration for Tepa — I've come to appreciate phonemic writing systems a lot more. Changes in my habits of language construction drive this somewhat, too. So, here's an account some of the considerations that went into settling on the Latin orthography for Kahtsaai.
Here's the Kahtsaai vowel inventory:
|i [i]||ii [iː]|
|e [ɛ]||ei [eː]||o [o] [ʊ]||ou [uː]|
|a [a]||aa [aː]|
The first issue I had to deal with is tone. I'm very fond of tonal languages — more fond than typology would warrant — but there it is. The only practical way to indicate tone is with diacritics.1 Since I stick with simple two- or three-tone systems, this is easy. In a two-tone system I use á for a high tone and no accent for low, and for a three-tone system á high, a mid and à low.
However, once I decide to use tone, I'm only really left with one option for long vowels, something else I'm fond of. In a non-tonal language, I use the acute accent for a long vowel. But, since I've already grabbed that diacritic for tone in Kahtsaai, I simply write the vowel twice to indicate length, a and aa, etc. (In the ancient times of ASCII-only terminals, that's how I always wrote long vowels.) In theory I could combine diacritics, and put accent marks above macrons, but I find that difficult to read and a real pain to write legibly or type. In Kahtsaai, each mora of a long vowel may have its own tone, leading to tone contours on long vowels, káar to save, to preserve having a falling pitch.
You will also note that the mid vowels aren't marked long in the same way. Phonemically, e and ei are just short and long versions of each other, but there was such a significant quality change that I decided to write them differently. This does work out in the phonological processes of the language. Noun stems that end in vowels lose a single mora at the end when they are incorporated. So, the noun kopi water becomes just kop- when incorporated, and éi tree has the incorporation form é-. This pattern also motivates the spelling of the single, long diphthong as aai. When final, the moraic reduction results in -aa, as in taraa- from taraai health, condition, status, weather. I think the switch from aai to aa conceals the stem less than a spelling change from ai to aa. The extra reminder that this is a long vowel diphthong doesn't hurt, either.
Finally, the phoneme /o/ has two realizations. In open syllables it is [o], in closed it is [ʊ]. The morphology of Kahtsaai ensures that underlying /o/ in a single root presents itself in both shapes frequently. For example, using the verb -wo to eat, te'ewo I ate it has no evidential due to the first person subject, and is pronounced [tɛ.ʔɛ.wo]. With the direct evidential, -ts, we get yonwots she ate it [jʊn.wʊts].
The consonants of Kahtsaai are much simpler. I decided not to follow the Americanist tradition of spelling /ts/ as "c", and just use ts. At morpheme boundaries t + s results in tss, so no ambiguity about stem boundaries arises from using this digraph. Since Kahtsaai allows coda stops, this could have become a minor problem.
Before voiced resonants (l r) or glides (w y) the stops (which includes ts for this discussion) are pronounced voiced. This change is not represented in the practical orthography, [kid.ɾa] to tame, subdue is spelled kitra. Again, this choice is motivated by not wanting the basic stem to be concealed in writing every time a new morpheme was added. Besides, the change is 100% predictable.
1 Ok, some languages use what look like coda consonants to mark tone instead of actual syllable codas. Hmong, especially, comes to mind. But I tend to favor moderately complex syllables, with actual coda consonants, so that could get very confusing.