Sorry, I find this too hard to follow.  At the end:

> ['TA', 'CA'] # Missing the substring 'CA'

the comment claims it's missing 'CA', while the output plainly shows it's _not_ missing 'CA'.

If your complaint is that's missing 'AC', yes, it is.  It's not intended to find _overlapping_ matches at all (read the docs).  The longest matching blocks in

TAGACCA
ATACA

are in fact TA, CA, and AC, but after the leftmost-longest TA is matched first, AC no longer exists _to_ match in the second string.  Only CA does.