Certainly.
The introduction of new technology like that, a new system to manage risk, is something that needs to be planned very, very specifically. In this case, it is true that in the first couple of years we were not good at collecting information in two areas. One of them was the specific results obtained by the machine; the other was creating a sample or another parallel area where, say, containers would have been opened randomly or people referred randomly, and we measured the extent to which we were doing this better than a random result.
We have done this now. Following the visit of the Auditor General, we now have a system that collects that information. Version four of our risk-scoring machine actually integrates that information; we don't even have to input it. It compares the results and gets additional information. Not only that, but a portion of it uses artificial intelligence to do something that no human being could do, to combine the risk indicators in different ways based on what you've been saying, based on past information and past results.
So that's significant progress, but it's true that at the first level, with the first package over the first two years, we didn't do enough of that. But you need that as well. You need a bit of a time period to justify what—