To be fair, perhaps it is lossless, but only for a subset of all possible data files. Like compressive sensing, which beats the Nyquist-Shannon limit, but only

To be fair, perhaps it is lossless, but only for a subset of all possible data files. Like compressive sensing, which beats the Nyquist-Shannon limit, but only if conditions about sparsity are met. In fact, the Nyquist-Shannon theorem itself is a kind of lossless compression. It says you can potentially throw away data without losing information.