As for the regexes, that’s a challenge since we don’t really know what bounces it will classify, you’d really have to build out a test harness for that kind of thing if it’s a concern, where it takes a list of bounces and runs them against the file, then lets you know which bounces didn’t get classified and which rules didn’t get hit.
Ultimately I’d say you validate the TOML, load it in on one KumoMTA node, check that your unclassified bounces in the logs aren’t excessive after a while, and load it to the rest of the nodes.