When we designed Coin, we assumed builds are 100 % reproducible. Reality bites us and we can have various external factors failing builds where when the same build is later repeated, it may succeed.
Currently all of this works, except that when re-running, we override the first build log. This is confusing, because we will see in the web UI for example that a build failed, but then the log says that it was successful.
Proposal: for testing, we have several test logs, when repeating a round of testing, we create a new file. Implement the same structure for building and provisioning.