Feedback and Questions

Yes, you have identified a missing extension to URs, not a bug as such. I thought I'd covered the compliment to UR Type 3, but wasn’t looking for cells in the same box. So I have added this to the solver and the documentation. Great catch!

Works for me on chrome and Win7sp1 x64 which I also have. Can you check it's not opening in a window or tab that is already open and/or underneath something?

I think it's safe to remove that 7. I'm pretty sure the way I've coded this the solver wont remove candidates it find in links coming out of a grouped cell. Probably being cautious. But logically there is no reason why its not a valid elimination.

I'd like to implement a full reverse history feature. The solver is designed to take any board at any stage of the solution process so it does not know when a puzzle is at the beginning or not. So there is no auto-restart. When entering clues you can tell it which are clues and solution but that only effects the color of the numbers. However, if you use it regularly you can use the [save] and [reload] before you kick off the solving steps. That will restore the puzzle. I usually keep puzzle as strings (using the [Email] button and copy/paste) if I'm working on several or many puzzles.

I've added a feature to the solver recently to help hunt for X-Cycles and other chaining strategies. Under the list of strategies is a set of checkboxes that you can use to highlight bi-location candidates (two left in a unit). If you suspect an X-Cycle you can check each number individually and perhaps build up a loop. You only have to alternate ON/OFF and ON implies an OFF link which doesn’t have to be bi-locational (since it removes all other instances). An X-Cycle has no start or end, you can construct it from any point, which makes it a little easier to spot. Not every X will be used because loops can be quite small (X-Wing is the smallest) but you can widen them to make them larger. Lots of X that never form part of a loop imply they are stranded in groups where they can't be linked in - there must be 3 or more in each unit. It is possible to have two or more isolated loops that cant be joined but this is rare.

I hope this is slightly helpful. Best is to look at examples found in the solver in the harder puzzles. They turn up a lot.

Are you talking about Killer Sudoku? It's the only context I can think of that might have a negative number in a clue. If so, then the solution to the cells in the cage could still be in any order or combination. There are only a very small number of cage combinations that fit minus numbers and only 2-cell cages can have them.

Interesting example, I'll take a look why this isn’t picked up. Thanks for sharing

Thanks for sharing it. Published for others to try and solve.

That behaviour is by design. The solver *could* check if the puzzle has only one solution but that would be a huge overhead especially as it has no memory of previous steps and this would have to happen every time someone presses "take step". So we assume the puzzle has only one solution. However, logic can only work if the puzzle *has* one solution. It will find some steps logically but ultimately it will get in a mess and create a contradiction. If you don’t think this approach is correct think about a puzzle with just one number - how would it 'logically' proceed?

Thanks for this email. I've corrected the message to say 12 clues, but I'm afraid the server will not return a solution count with so few clues. I have to throttle the services since it's a busy machine and the brute force tester sucks CPU. I can confirm my offline solver finds one solution. I have an article about this puzzle.

I'd be interested in the first move you found for Grouped X-cycles. Would be nice to make further progress in that area.