It looks more like Google execs losing their mind and pressuring researchers to put DeepSWE directly into the training set.
It's clearly been "dealt with" already. When it launched we had interesting gaps and definitely differences. Now every new release is "crushing it".