I'm looking back at some old questions from the beginning of the site, and I've noticed a set of questions that, to this day, are still posted.
It started when I re-viewed the Security to the Party series. Looking through the list of all 39 questions, I'm struggling to find any of them that are of high quality. I'd ideally like to put (nearly) all questions in this series, and potentially also of this style, on hold as too broad. Looking through recent examples, it's very difficult to find any that are done well enough to limit the scope of reasonable answers to any degree - there are always many detectable patterns, as the sample sizes are just too small.
In my opinion, questions where the asker comes up with a pattern and provides a couple examples of them, and we're supposed to infer backwards off very little information what the original pattern is, are categorically too broad - at least how they're done right now.
Compare these with the "What is a _____ word™?" series, where each question had 10-20 examples, and frequently well over 15 - and the only reason I think these worked at all is because there were actually that many examples. With fewer examples, more wrong guesses would start to match all the provided clues. Only one password puzzle has provided a comparable number of examples.
Under what circumstances, though, are these questions not too broad? Can you cite, with examples, situations under which they become sufficiently well-defined?