Just to clarify, since I misread your quote and think others might as well:
> "The replacement Kerl hash function is unmodified KECCAK-384 that only converts its input and output from/to 243 trits to 48 bytes using basic two’s complement. KECCAK-384 is well vetted and researched."
"Replacement" really does mean replacement, i.e. what they substituted in after the researchers successfully attacked their original hash function. (The new one is called "Kerl" and the old one is called "Curl", which seems confusing.)
I'm not sure how asserting that the replacement is well-researched addresses the OP's point that their history of (a) rolling their own hash function, and (b) not adequately considering differential cryptanalysis, is a red flag. True, that specific vulnerability can now be considered fixed, which might not be clear from the OP's post - but it's still evidence of incompetence, which doesn't bode well for the quality of the rest of their design.