Feedback and Questions

Sure you weren't looking at a Sudoku X puzzle? That is where the diagonals have a uniqueness condition
http://www.sudokuwiki.org/sudokux.htm
However, some normal Sudoku can have a solution with uniqueness in the diagonals, but its no help unless you know that - which you can't until the puzzle is completed

That is weird and has been pointed out to me recently. I know I’ve updated the Killer stuff in places over the last year, but it should not be two steps forward, one step back. Nothing *should* make an old puzzle harder or impossible since it was created. Generating and solving a puzzle are two sides to the same coin. Something is not quite right so I'll have to find time to figure it out.

Eliminations are correct. That is a 2-3-2 swordfish based on the missing 5s in those cells. Not necessary for all 9 cells to have the common candidate, have a look again at
www.sudokuwiki.org/Sword_Fish_Strategy

Not a bug: that’s expected behaviour. The way Solution Count works is to insert every possible number into the grid starting at the top left, working the row and then going onto the next row. To make this work in a second or so there is a fast way to block potential dead ends and reduce the search space. Each insertion removes that number from the rest of the board according to the Sudoku rules (unique in row,column and box). Backtracking up a failed path restores the numbers. How large the search space is depends on the density of clues at the start of the search. Ie, few clues in the top left = much longer time. Rotating the puzzle will change the density. There are many symmetries in Sudoku in addition to rotation and one of them will be optimal for this algorithm. I would like to add a check at the start to decide which rotation would give the best performance. Any correctly implemented symmetric transformation will still give the same number of solutions - and the same grade and solve path (except for those strategies which also have a top-left/bottom-right search bias)

Hope that’s interesting

It’s a hard fact. They are different sides to the same coin. Applies to Singles, Triples, Pointing Pairs / LBR
They produce a Naked set after eliminations. Where a Naked set goes onto remove more candidates will only occur in another different unit. For example a Pair might be uncovered in a Box but then the same new Naked Pair can go on to remove candidates in a column (outside the box). This happens a lot in Jigsaw Sudoku

Good point. I've just been working on calibrating grades for the new Windoku puzzles, so I've taken a look at the 'tough' puzzles in the Daily Sudoku list. I can indeed see about 15% which are using extremes and about 5% which are too 'boring' in having very little strategy stuff at all. There is a warning flag for both these possibilities so I've used that again on the tough set to replace the ones with the warnings with new ones that don’t. So overall I think that grade will have less outliers it in now.

OK this works out this way: The 1/2/3/4/5/6/7/8/9 is built up of all the combinations that make 13 with three cells. That is literally true. Hover your mouse over the cage in the small cell. But there is only one combination that includes 9 - {1,3,9}. There is only one cell with a 9 available - B2, so 1 and 3 must go in A2 and A3. But there is no 3 available in any cell, so this combination isn't available, hence 9 can be removed from B2.

I can see the virtue of expanding the result explanation to explain this, tricky to generalise

W-Wings are encoded in the solver as an experiment/investigation into a larger family of wings. But I've found they duplicate other strategies and I've decided not to release them. Their adherents claim these are all unique patterns worth looking for but I think they are formations of a more general wing-type and don’t need to be broken up into these family types. I'll try and get a page up at least of w-wings. Google might have some example on various forums in the meantime.

Did this W-Wing appear in the solver?

Thanks Brian!

Kakuro is a nice puzzle, one do make them and have an offline solver, in that sense. But there's not many different strategies to it, it's pretty much combination elimination all the way. I've thought about it but I'd like to get more newspapers taking the puzzle before I put the time into it. Happy to be pointed at examples that are interesting though. Also, Kakuro can come in any size and shape making it very awkward to design a solver interface.