Every integration estate has one. A point-to-point interface built years ago by someone who has left, moving a file nobody can fully explain, which three downstream systems now depend on. It has no owner, no documentation, and no monitoring, and the reason it survives every rationalisation programme is that nobody can prove what breaks if it stops.
The usual response is to leave it alone and route around it. That decision is rational each time it's made and expensive in aggregate, because the estate accumulates these until the cost of change is dominated by fear rather than effort.
There is a cheaper way through it than a full discovery programme.
Instrument before you investigate
You do not need to understand the interface to observe it. Log every invocation for thirty days: who called it, what came out, what consumed the output. Most estates have this data already in gateway logs, file-transfer audit trails or database access logs, unaggregated. Thirty days of observation costs a few engineer-days and answers the question that months of documentation archaeology usually doesn't: is anything actually using this, and for what?
A meaningful proportion of "critical" undocumented interfaces turn out to serve a report nobody reads, a reconciliation that a newer system now performs, or a downstream consumer that was decommissioned and left its scheduler running. Those you can retire on evidence rather than argument.
For the ones that are real, write the contract before the code
The reason the original is unmaintainable is not usually its implementation. It's that the agreement it encodes — what fields, at what frequency, with what quality guarantees, and who is accountable when it fails — exists only inside the running system. Extract that agreement into a written data contract with a named owner on both sides. Once that exists, the implementation becomes replaceable, which is the actual goal. Teams that skip this step rebuild the same fragility on newer technology and discover it eighteen months later.
Then decommission loudly
Turn it off in a controlled window with the consumers watching, rather than quietly with a rollback plan. A failure you caused deliberately at 10am on a Tuesday, with the right people in the room, costs a fraction of the same failure at quarter-end.
The estates that stay maintainable are not the ones with the best architecture. They're the ones where somebody had the authority to remove things, and evidence strong enough to use it.