What are Insular <tw b y> from? *tʷ *kʷ *jʷ, presumably.
The original plan was for a chain shift of v w > w ɥ, but this doesn't seem reasonable. The lateral isn't terribly common in Kett, so maybe it developed later and never had a rounding contrast. *t *k *j being the only consonants with a rounding contrast doesn't seem reasonable, though.
In macro terms, it's possible that the rounded plosives developed from something else and the rounding distinction developed later - cf. Marshallese, where /tʲ tˠ/ are from *t *s but some Cʲ/Cˠ/Cʷ pairs are just from vowel transfer.
Other rounding contrasts could've been lost. Northern Sangtam has /tʙ̥ tʙ̥ʰ nm/, which presumably once formed a series, with AFAIK no evidence of rounding elsewhere. So rounding would've only been on coronals. These eroded asymmetrically, into /t tʰ ⁿd~ⁿb/. So a reversal of erosion seems plausible, with *nʷ merging into the reflex of *n. Adding unit /nm/ to Insular doesn't seem objectionable, but it could be lost, as could /ŋm/. /f/ is presumably from *xʷ; /ŋ/ could be from a merger of *m and *g, or something like that, with *gʷ > m or mb or whatever. /ŋ/ is very common.
The most reasonable candidate for the immediate ancestor of Insular is probably something like:
/t tʷ k kʷ/
/b d dʷ g gʷ/
/θ θʷ x xʷ h/
/m n nʷ/
/l (lʷ)/
The original plan was for a chain shift of v w > w ɥ, but this doesn't seem reasonable. The lateral isn't terribly common in Kett, so maybe it developed later and never had a rounding contrast. *t *k *j being the only consonants with a rounding contrast doesn't seem reasonable, though.
In macro terms, it's possible that the rounded plosives developed from something else and the rounding distinction developed later - cf. Marshallese, where /tʲ tˠ/ are from *t *s but some Cʲ/Cˠ/Cʷ pairs are just from vowel transfer.
Other rounding contrasts could've been lost. Northern Sangtam has /tʙ̥ tʙ̥ʰ nm/, which presumably once formed a series, with AFAIK no evidence of rounding elsewhere. So rounding would've only been on coronals. These eroded asymmetrically, into /t tʰ ⁿd~ⁿb/. So a reversal of erosion seems plausible, with *nʷ merging into the reflex of *n. Adding unit /nm/ to Insular doesn't seem objectionable, but it could be lost, as could /ŋm/. /f/ is presumably from *xʷ; /ŋ/ could be from a merger of *m and *g, or something like that, with *gʷ > m or mb or whatever. /ŋ/ is very common.
The most reasonable candidate for the immediate ancestor of Insular is probably something like:
/t tʷ k kʷ/
/b d dʷ g gʷ/
/θ θʷ x xʷ h/
/m n nʷ/
/l (lʷ)/