I recommend that potential contributors start by
getting to know Data.CompactSequence.Stack.Internal.
The basic operations there require considerably
less code than those for the other structures, so
it's easier to understand how they interact with
the debit invariant. The other structures work
according to exactly the same principles, but with
a lot more moving parts.
Speaking of debit invariants, contributors who want to really dig into the deep guts of these implementations should be somewhat familiar with Okasaki's debit analysis framework. This is explained very accessibly in his thesis, as well as the textbook he expanded it into. There's no need to fully understand his justification for the framework; just get a sense of how it's applied.
Note that almost everywhere in this package, we either must be lazy or must be strict to maintain the desired space and time bounds. There's a little wiggle room in some places, but not terribly much.