An existing application may be valuable even when its original developer is unavailable. It can also contain undocumented dependencies and business rules. The first task is to establish enough truth to make a safe decision, not promise a fixed-price repair after seeing one screenshot. An investigation should have a cap, a useful output and clear stopping conditions.
Find the actual operating baseline
Ask who owns the source, licences, server accounts and data, and who can authorise inspection. Record the running version, supported stack, integrations, scheduled jobs and existing backup/restore arrangements. Compare the source available with what is actually deployed; an old repository is not proof it can rebuild the live application.
List known faults and the requested change separately. “Add a report” is different from “make the entire system reliable”. Capture examples of current behaviour using minimised, authorised material. Avoid copying confidential records or credentials into public tickets, source or external AI tools. A customer’s staff member may not have authority to grant access to a supplier’s platform.
Make the investigation commercially bounded
A useful discovery result identifies the likely change boundary, dependencies, risks, options and remaining unknowns. It may recommend a focused patch, supported product replacement, phased migration or no intervention until a recovery gap is fixed. It should not simply consume an open-ended number of hours with no decision point.
If work proceeds, agree acceptance criteria and regression scope around the relevant business paths. Record which pre-existing defects are excluded, while retaining responsibility for new failures caused by the change. Treat supported versions and third-party retirement as explicit dependencies. A broad “legacy system” disclaimer cannot excuse careless modification.
Test the recovery path before the release
Reproduce the relevant behaviour in a safe environment where practical. Establish a recoverable source/configuration/database baseline and rehearse the change. Test both successful and denied/error paths, and define a rollback decision. Some data changes are not reversibly solved by switching a source pointer; the plan must account for real writes and dependent jobs.
Customer approval of a proposed solution, engineering test completion and permission for a live release are separate decisions. Bind the release to a precise version and target. If the running estate has drifted, stop and reassess rather than applying a patch to a different system because the filename matches.
At handover, preserve the newly learned operating knowledge: dependency map, source identity, test evidence, remaining defects, recovery instructions and maintainership. The best first improvement may be making the system recoverable and understandable, even before the requested extra feature is delivered.
Sources and review
Prepared by Yellow Brick Tech. General guidance; the source and test conditions for your own situation still matter.
Next review: 4 May 2027