Following the Gumuz tangent by looking now a bit more into Koman. Conveniently, for this there is a recent dissertation “A Historical Reconstruction of the Koman Language Family” (Otero 2019) already sketching out all the main lines of development.
The family looks phonologically quite compact really, but low on cognate vocabulary: 25% Swadesh list counts between the two main branches and 20% with poorly attested outlier Gule. Probably though this is not really due to the family having an age of 4000+ years as given by glottochronology estimates, when the sound correspondences are mostly transparent. My spitball take would be closer to 2000 years old and heavy relexification due to low sociolinguistic status. Per Otero, the Koman speakers have apparently for long suffered of slave raiding and land theft by their bigger neighbors. For that matter, maybe loanwords from adjacent languages could explain why the family has been for long thought to be a part of Nilo-Saharan? That other 75% of basic vocabulary has to come from somewhere.
Shoutout in particular to Gwama for showing multiple interesting sound changes:
word-initial *ɓ *ɗ > /pʼ tʼ/
tone-conditioned devoicing: word-initial *b *d *g > /p t k/ before low tone (however *ɟ > /z/ always, without devoicing to /s/)
in-progress word-medial chainshift with /b d g/ > [β ɾ ɣ] and /pʼ tʼ kʼ/ > [b d g]
*s and *ʃ merge before high vowels according to ±ATR: as /s/ before /i u/, as /ʃ/ before /ɪ ʊ/
The tone shift (also found in some other Koman languages) is actually a sort of cheshirization, as low tone is itself conditioned by *voiced initials. Not immediately though, since the tone split is attested also in languages still retaining voiced stops (and also the devoicing also in some languages that appear to have re-merged the low and mid tones).