From the wording of the description of the programmer who failed to debug and labeled it user error it appears that it is fairly typical Accenture-grade software where there is no single bug so much as the program itself approximates the correct result.

Their data model appears to have been akin to having a single accumulator sum up things rather than to use something like double-entry bookkeeping or an account graph so that the source of errors could be traced.

It’s less “a bug” and more a coincidence that the application worked when it did.