News with judgment

The testCan a decision model rank a live feed against a profile you wrote in two lines: no keywords, no training, no examples? Each headline is one call with the same four questions; your profile travels along as part of the state.

How to read itThe big number is interest, 0 to 4: the expected value over a five-level rubric, so 3.1 means the probability mass sits between "quite a bit" and "a lot". Cards rank by it. The red tag is the winning topic with its probability; clickbait and factual are plain yes/no probabilities.

What to watchDoes the order match what you'd pick? Do the things you said bore you sink? Open "all answers" on a borderline card and look at how spread the probabilities are. Then check the strip: latency per call and what sixty headlines actually cost.