The primary goal is to make Releng work more transparent, by tracking everything Releng produces (composes, update pushes) across all releases (current, pending and Rawhide), and all architectures. (primary and secondary)
We want to be able to easily tell, at all time, what exactly was produced. (the NEVRA of all packages, the keys used to sign each one of them, the size of the isos,...)
Knowing how each was produced would also be very good to have. (the versions of the compose tools, the kickstart used to create it, ...)
Finally, we want to be able to easily diff two composes, to quickly identify changes. (for example, to figure out why TC2 is oversized while TC1 wasn't)
The current process to compose Fedora uses the run-pungi script. This script could be made to upload its results to the Compose DB.
QA could request Test Composes through a proper form in Compose DB rather than through a Trac ticket. This could even start parts of the process automatically. (up until a human needs to sign?)