"I have thought about the particular challenges of this year's Advent of Code (which I greatly enjoyed once again, with some intermittent mixed feelings), and my conclusion is that some problems might very well be of similar difficulty levels, but with nevertheless very dissimilar frustration-inducing levels. And I think the frustration level is strongly correlated with the feedback signal one can use to debug. Case in point: Day 15, which was almost impossible to debug, once you had a working solution, versus Day 17, which was actually fun to debug, since the visual signal you could use was so strong."
Ultimately though I realised that if I followed the description word by word, and made no assumptions, then I got the right solution. Still, a nightmare: you could pass all the examples, or even some alternate real sample data, and still get it wrong if you got the ordering slightly incorrect.