What if a team maintains a long-lived legacy system, where critical system knowledge is implicit, embedded in manual regression tests and in the heads of a few experienced team members, some of whom are approaching retirement? Arthur van Leeuwen and his team faced the risk of losing that knowledge and decided to counter it.

This talk explains how Arthur and his team captured implicit system knowledge from their 25-year-old legacy application before their regression testing expert retired. Their regression test suite covered essential system behavior, but its intent was difficult to understand without years of accumulated context. Rather than focusing on test automation, they used Gherkin as a collaborative language to make that implicit knowledge explicit and accessible.
They started translating their manual regression tests into Gherkin following commonly accepted best practices. It quickly became clear that generic guidance did not suffice. They wanted to specify behavior for human understanding, not for writing automated tests.

Together, the team had to figure out what “good” specifications looked like for them. This required learning as a team, as well as individually. Furthermore, it meant looking beyond development and testing, and also involving documentation specialists. They agreed on terminology, partly derived from their user documentation, and on rules constraining how they formed sentences in Gherkin feature files.

The resulting specifications reduced reliance on individual experts, enabled non-experts to perform regression testing with confidence, and exposed underdocumented behavior and previously unknown bugs. More importantly, they became a practical communication tool for the team. As a bonus, they also proved to be a good foundation for test automation. In this talk, Arthur distills these experiences into practical insights on how and why to introduce Gherkin as a shared specification language, define effective wording agreements, and use those agreements to make implicit knowledge explicit in long-lived legacy systems.