Stranger in a Strange Land - Croco-puzzle review
One thing to start such ratings out, at the individual puzzle level, is simply time goals applied to printed puzzles. Nikoli does this on some of their puzzles (more types online than in print) and beating the "expert" time is the goal I tend to have when solving on paper, and I note if I beat it by 2x or 3x or such with double or triple circles of the time. Sudoku is hard to rate, since most computer solvers don't approach the puzzle like a human does, but Wei-Hwa experimented with writing code to do this before the WSC and his results seemed pretty good at mapping human times. I'd like to get back to advising him on that sometime soon, and then implementing it in a printed book or two so that solvers could solve a classic sudoku and know, based on a percentage of a goal score, how well they are doing relative to other solvers.
But on the broader topic of rankings, with so many online sites and live tests, there is not much out there. A long time ago Cihan Altay tried to keep this up, using live and online tests for rating points, to give a top list, but this stopped just as I was starting to compete. Right now, Logicmasters India is making rankings based on their sudoku and puzzle tests. I happen to be leading the sudoku list despite sitting out many tests (I simply don't love sudoku the way some do) and am tied at the top of the puzzle list right now with Hideaki Jo; Ulrich Voigt is in third after a poor Flip test - otherwise he'd probably also be essentially tied with us since we are often seconds apart on an hour plus test. (Aside: an excellent "Puzzles and Chess" competition by Nikola Zivanovic is running this weekend and will be open for another 5-6 hours so do head over there to print out the puzzles for later if you can't find time to compete). Even these rankings require solving one or two hour tests on weekends and don't extend to other settings or other sites well since the formula at play is hardly transparent.
Well, the German site Croco-puzzle was brought to my attention by Stefano. I've avoided the site for awhile because A) there is a huge language barrier since it's all in German, B) I suffer from puzzle-snobbery, the puzzles are computer-generated and many, like a CG Hitori, leave me with as much "joy" as you'd expect, and C) a general dislike of online solving of any form, because applets rarely give me the freedom paper does to notate puzzles in a good way and I want to practice for live competitions, not learn bad habits from a particular applet.
However, Croco-puzzle has a very interesting ranking system that is (relatively) easy to understand and seems to be a good basis for a puzzle league/ladder system. Basically, every day solvers get access to a puzzle (since Dec. 1st, two puzzles), which they solve. After the day is over, all the times, including solvers who opened but didn't finish the puzzle, are considered to calculate a median time and a best time. The best time is worth 3000 ranking points. The median time 1500. Anything else is scaled based on its position relative to these two times through a formula given on the main page. While this can give a score for one day, the way to make it a true ranking system is that only a fraction of that day's points counts for your next ranking. 59/60ths of tomorrow's ranking will be equal to 59/60ths of today's ranking. 1/120th will be equal to the score on that day on each of the two puzzles, from 0 to 3000 depending on your result. Everyone's scores go up and down. If you are successful at a certain score level for long enough, you earn a new rank much like in judo or other systems, and rankings exist at every 100 point plateau so everyone has goals to shoot for both personally and competitively with others. If you are away for a week, your rating won't change at all. Only days you compete will ranking points increase or decrease.
After Chris Dickson of the UK made a nice walkthrough of some of the site in English which brought it to the fore of my mind right after the WPC again, I thought I'd try it out for a month. I didn't want my trial to be publicly known, so I did not register as motris or drsudoku as I would elsewhere. I wasn't sure I'd stick with it, and the mystery of who is this guy could be fun. So instead, I made a half-hearted attempt at a new nom. In a cryptic crossword sense, a half-hearted attempt is exactly what I did as half of my middle name is MARS. Also, solving on a site exclusively in German is like being a solver from another planet, so MARS fit pretty perfectly. The first week didn't go so well, as I was at best barely getting within 10% of uvo times and the puzzles were new and frustrating (a Pyramid I really stumbled on, an ABC-Box that I just didn't solve fast, an insanely large and un-fun Domino "puzzle" that took me almost an hour to hammer (not logic) out that may now be known as the Refrigerator puzzle as someone took delivery of that appliance during the puzzle and got credit for it alongside their time). Aside from unfriendly or unfamiliar puzzles, a large part of my stumbling was in learning the new applets which are not easily documented as their text can't be pushed to a translation engine as easily. Many of the applets have hidden buttons of incredible value. Dominoes has a "D" that marks out a full domino. Didn't know that when I first tackled the Refrigerator puzzle. Magnets had, at the time, an "F" key that fixed all the magnets unknown around a particular one to match polarity. An awesome feature, if you know to use it. New updates to the applet have changed some things (Magnets now does "auto-F" compared to the old setting) and also given a good trial and error feature that can instantly backtrack, a good addition to the four colors that can be used to stratify guesses as well. Some have inconsistently applied systems. The "#" route to sudoku notation is too cumbersome to use. I prefer the "-" based double noting: 1-2 puts both numbers into a cell. This works on Killers and Kakuro, but doesn't on regular sudoku. So remembering where notation can and can't work is an extra step some days. Some of the applets work quite well. Others, particularly loop-drawing ones, are nowhere near as easy to use as the Nikoli ones, and even Nikoli's can't suffice over the ease of drawing on paper.
Many days I enjoy the puzzles greatly - they have a quite varied spread and sometimes feature tough forms of some favorites like Skyscrapers and Easy as ABC and Star Battle and Tapa and other enjoyable WPC types that simply aren't as easy to find elsewhere regularly, and certainly not in a time-based setting where you can see how you stack up to others. As noted above, Magnets (with the applet assists) is a favorite for me on Croco-puzzle even though uvo will top it most all the time - but maybe "auto-F" is a bad habit to gain before the next paper test. Other days the puzzles aren't what I'd want them to be, almost exclusively when they venture to Nikoli types. The Slitherlink (Rundweg) on croco-puzzle are actually pretty good, particularly as they use large boxes or hex variants regularly, but the rest range from average (Masyu, Heyawake) to poor (Arukone = Numberlink, Hitori). The Heyawake generator, for example, is in love with 1 x n rectangles clustered together, as well as 0 clues being seeds, which means you can quickly adapt your mind to how the solutions will tend to look/expand. The applet on Heyawake also doesn't fully shade cells which I dislike compared to Nikoli's presentation. The sudoku generators are very hit and miss and I hate the inequality ones, particularly on the harder end. I made thermo-doku to improve inequality sudoku puzzles, and these rub me the wrong way as one would expect, particularly as the forcing nature of the constraints is often way too subtle for a puzzle. If I need to bifurcate to show something on a sudoku, the puzzle is broken (for a human solver).
The "new" puzzles on the site also split across a range of like to dislike. Some really don't feel like puzzles to me. Sternenhimmel feels like "work", just like a Hitori does. Basically, if I had a button to remove all the unpointed at squares, this is almost never a puzzle. If removing the unduplicated numbers similarly ruins most Hitori, and certainly these, you can understand why I don't enjoy either as there is no thinking to be had and a puzzle should engage the mind. Pillen sits on the other end of the extreme, with some search elements and logic elements that can give a good challenge. While the initial grid looks like a Hitori, the solve is so much different and quite fun. Others are still very unfamiliar, but as I solve more of them, I've gotten better at some. Solving all the preisratsels (Prize Puzzles) from the last 8 years will very much train you in the various applets and how to solve and guess on croco-puzzles.
So, the first week was rocky, but then as I said with some practice I started to top puzzles. Then top some more. This felt a lot better to me than being second or third. The reason is that when you don't have as many points to defend, and will certainly increase each day, what you really want to do is get the top time and set a standard that reduces the number of points others get. This is very much about "playing offense", until you have ranking points to defend (I'm approaching that point finally). So I aggressively went after puzzles as fast as I know how to, on some types this meant quicker bifurcation when I really didn't need to do so, and doing less checking than I might otherwise. Sometimes this costs me a lot (like on a laser recently, or a Hitori I would have beaten everyone by 40 seconds on), as errors are very bad. Unlike sites like Nikoli.com, where an error just makes you go back to the puzzle with the mistake highlighted, here an error affects your time with a penalty equal to the median time of all solvers without errors. This almost always means you are going from potentially a max of 3000 to closer to the 1500 range or worse. The red text that you've made a mistake is a real stress inducer. And I have in my haste made the common errors I do on some of these types. A Tapa had an unconnected white cell. New approach: mark all unused cells too to check. A Hitori had an unmarked black that was needed to not have 2 solutions but otherwise unimportant. New approach: mark all unused cells too to check. I'm adapting and getting better.
And mars got noticed and exposed. Sooner than I wanted. Berni actually announced who I was in week 2 of the experiment on the German forum, long before I could build any anticipation of what new solver is setting benchmark times on ~20% of the puzzles aside from a single UK forum post asking who I was that went unanswered. As it stands, after 36 days, I have solved every puzzle I've seen (48 now), topped 9 (in place to get 2 more today though for 11/48) and have built my rank to 238th (1175 points). It will likely take about 5 more months before I will be in the 2500+ ranks, and longer to earn a 5th or 6th Dan rating, but while I find this slow change frustrating at the moment, it is exactly the right thing to do to have a fair and slow to change ranking. Puzzle ranking is absolutely done right on this site.
How would I compare this to the only other online place I play (Nikoli.com)? Well, I wouldn't pay for croco-puzzle daily without some changes, but I don't have to as this site is free to register and play on. The ranking aspect of Croco-puzzles is so much better from a competition stand-point than Nikoli. On Nikoli, the only obvious ranking to go for is first to finish, since solvers will retake puzzles (and those scores then crowd the top of the time-sorted rankings for a reason I'm still not sure of) and solvers may not get to a puzzle for days/weeks so knowing your final spot requires looking back. There are no high scorers lists, for different puzzle types, on Nikoli.com, while this is a feature on croco-puzzle with all-time and current leaderboards. On Croco-puzzle, the surprise puzzle ranking is fixed after 24 hours, since solvers just have any time during that day to solve the puzzle. You can retake the puzzle as many times as you want afterwards (although they eventually disappear after 48 hours entirely from the site), but the ranking list is set from the first time, and is easy to follow and watch from day to day. The goal of "double-top" has been my attention for the last 11 days, since the double puzzle came out. Hausigel (Roland Voigt) did it first, but I might squeak it out today with an impressive Heyawake solve and a 1 second faster than misko (Michael Ley) Buchstabensalat (Easy as ABC) solve.
Even aside from the rankings, the range of puzzles is great WPC practice that simply isn't available doing just Nikoli.com. So, to catch back up to Ulrich I'll be playing here going forward. I would highly recommend other putative team members try it out too, maybe reading through Chris's walkthrough in English to get a little sense of what does what, but the experience for now is still very much like a stranger in a strange land. Everything will be in a foreign language, and until you are sure of a type and how the applet behaves for it, you might not be comfortable solving it. So certainly not where I'd point beginners, but a good experience for those who want to really figure it out. The site has a lot more to it than the surprise puzzle and rating system - the old Advent puzzle sets have some really good tough puzzles on the highest levels - but the rating system alone sets a model that begs to be copied and used elsewhere in the puzzling world. Using Nikoli-quality puzzles with this rating system would be exactly what I'd implement if I ever made my own site. Then we could establish who is the best across each of several puzzle types, as well as overall, in a compelling way.