Generally the same, but I don't think unique word count was greater than 80. For example Aug 2015 #1 and #2, different sizes and layouts but almost identical word sets.
By the way clicking 'Download Image' on any crossword results in an image called 'Dec 2015 (#0) for N5_ Kanji -_ kana.png'. Every image has an identical content.
Sorry - Internet was out during a move to a new house. It's possible I was looking at different lists from you, but I had seen more variation in them after the rebuild (rebuild meaning that I erased all the recent new ones, and told the site to make a new set of crosswords with the same names). If you can tell me which ones you are looking at, I'll see if I can't fix it up (and quickly this time).
Look at all the crosswords from this month. There are 11 crosswords ( September #1 to #11 ) and if I didn't make any mistakes they use only 50 unique words. 18 of those words appear in every crossword. I think 50 words isn't enough to make those crosswords feel unique. Here is google doc with more data: Google Doc :)
Thanks for the extensive data. I've dug into it even more, and I believe I have the culprit.
The N5 level has roughly 500 or so words in renshuu's lists. Once you remove words that cannot be in the crossword for one reason or another (they don't have any kanji, or no kanji at the n5 level, they are katakana, etc), it drops down to 204 words. After removing words 3 characters or less, you get 80 words. The site randomizes them, but does a bit of guesswork to see which terms are "most eligible" for the crossword. Simply put, it looks at the characters in the word and sees how "crossable" it is - how easy it should be to intersect it with other terms.
Given a much larger list, you have enough randomness amidst terms with relatively similar scores, but not with numbers this low.
I've relaxed the rules a bit, and pushed it towards accepting smaller sized terms from time to time, but the base pool is still rather small.
I'm thinking it might need to be readjusted so it goes from easy > hard puzzles, but that would require breaking away from the JLPT levels, and making them much harder for those studying the JLPT (the majority of the site users).
It might also be a better idea to do what I do with the kanji crosswords, and just wipe out n5 altogether due to the insufficient pool from which to draw words.
