At first I thought this was a machine learning project that sampled news over a period of time and rejected nearly-identical articles that pop up over the same

At first I thought this was a machine learning project that sampled news over a period of time and rejected nearly-identical articles that pop up over the same amount of time. "important news! bill gates gives 5 million to fight COVID!" that pops up on 50 news sites a few hours apart could be rejected, while a lone article that only appears in one or two places would appear as non-trivial. Apply filter on HN articles and you get just the unique, non-trending stuff. Sample over 24 hours and you have the most popular non-trending articles.