I bet the only complete specification of the algorithm was the program running on that one machina, and that it was written in some ancient VAX assembly.

Also, $1M is a lot of money to risk subtle changes to the logic of a 40-year old program. Main reason why big financial orgs are still running software from the 50's on mainframes, that.

If it ain't broke, don't fix it.

I have seen first hand that some algorithms can be quite sophisticated and even having the source code (without documentation) isn't nearly enough when the binaries that support the mathematics only run on an archaic piece of hardware.