You must modernise a mainframe application with no documentation and one remaining expert who retires in nine months. Where do you start?

Treat the expert as the critical path

Nine months of one person's knowledge is the constraint on everything else, and it depreciates the moment they leave. Everything below is sequenced around that.

Record rather than interview. Sit with them and walk through the system, screen recording and transcribing. An interview produces what they think to say; a walkthrough produces what they do, including the steps they no longer notice they take.

Prioritise the undocumented decisions, not the code. The code can be read. What cannot be recovered is why the batch runs in this order, which conditions the overnight job assumes, what the workaround in the third step is compensating for, and which apparent bugs are relied upon by downstream processes.

Have them review what you produce, so errors surface while they can still correct them.

In parallel, extract what tooling can

Static analysis for call graphs and data flow. Job scheduling definitions for the batch topology. Access logs for which programs actually run — expect a large fraction to be dead. Database schemas and the interface inventory.

This is grinding work and it is the only source of truth that survives the retirement.

Then decide the strategy from evidence

The choice between encapsulation behind APIs, re-platforming, rewriting and replacing with a package depends on facts you do not yet have: how much of the system is actually used, how much business logic is genuinely unique, and how many integrations exist.

Resist choosing before the inventory exists. A rewrite decided on the basis of "it is old" is how a three-year programme starts and does not finish.

The interim safety measure

Whatever the eventual strategy, put a facade in front of the mainframe early. All access routes through an interception layer that initially forwards unchanged. It has no behavioural effect, it is safe to deploy, and it is the precondition for extracting anything incrementally later.

What to tell leadership now

The retirement is the risk, not the technology. Ask for the expert's time to be protected — a substantial share of it, formally — because every week spent on business-as-usual is knowledge that leaves permanently.