Should I analyze these consonants as codas or onset clusters?

Question

In my conlang, the proto-language had a CVR (R standing for a resonant) syllable structure and no voiced obstruents. Later, unstressed vowels were lost between voiceless obstruents (i.e. all obstruents). After this, obstruents were voiced when between two other voiced segments (/s/ became /r/, which was not present in the protolang), and all clusters with any voiced segments were split by an epenthetic vowel /a/.

The result is that the new language appears to be CVC, but only permitting voiceless obstruents to serve as codas, and only word-internally before other voiceless obstruents. Which made me wonder: does it make more sense to analyze this syllable structure as CCV? This doesn't end up working exactly (the protolang had uniform initial stress, preventing words from starting with clusters), but it feels like it makes more sense. So how should I analyze the syllable structure?

Answer 1

As already noted in the question, the answer is: It depends on the analysis. However, there are some tools to inform the analysis, most notably the sonority hierachy. The sonority sequencing principle describes a syllable as a sequence of sounds with peak sonority at the syllable core (a single vowel or a diphthong) and falling sonority towards the beginning and the end of the syllable. So, a syllable boundary is at the troughs of sonority. It is not necessarily uniquely determined by this requirement and there is another heuristic often invoked, the maximum onset principle: Put as many consonants as possible to the second syllable. This is not the only possible way to determine syllables, Ancient Greek grammarians used the heuristic to move the maximum possible word initial consonant clusters to the second syllable and Greek has a lot of interesting consonant clusters in this respect, e.g., /mn-/, /pt-/, /ps-/, or /kt-/.

Applying it to the conlang described in the question, we arrive at some words with the structure CVPPVC where P stands for a voiceless plosive. Since all voiceless plosives are on the same level of the sonority hierarchy, a syllable boundary between the two plosives is the choice suggested by this analysis. Putting the syllable boundary before the consonant cluster is not completely outlandish, compare the Greek words starting with /pt-/ or /kt-/, but the described conlang still does not have word initial consonant clusters, so it looks unnatural to me.