You are quite correct, I wasn't handling the situation where a user entered a valid set of numbers (that conformed to the rows/columns rule) but didn’t match the clues. I've re-worked it now so that it checks values against clues when the solver on the server is called (after the basic strategies) and once more at the end when every cell has been filled in. Your example now comes up with messages.

Refresh the page a couple of times and make sure you can see version number 2.08

Fixed! I didn’t quite get the data import right yesterday. Thanks for the alert

This is an interesting issue. Does that fact that we are setting +3[D4|E4] imply we cannot set any other 3 in the column to ON as well? If it does then whenever we make a chain we must cross out all X in every row, column and box that is used as we construct the chain. This changes the chain from a chain into a 'net' and it would be arbitrary to stop looking for the effects of such an ON at one step. One would have to have a full tree branching search - Bowmans Bingo works in this fashion.

I have built into the algorithm for chains the rule that the same candidate can't be used twice - mainly to stop endless loops appearing. But no rule exists for this type of check. However, if there was a problem doing this then I would have found an example where it creates a mistake. In the zillions of tests I've done this situation hasn't created a false positive, so I conclude it is harmless. I think the way to look at it is that these candidates are being turned on "potentially" - they have no reality to them since we don't know if they will eventually be the correct solutions to those cells.

Do you think there is always an alternative chain that doesn't violate this rule?

I am making some new stock for 2014 and I'm going to make enough to find examples such as these. I'd like to get about twenty representative puzzles that really highlight a particular strategy and don’t rely on any hard strategies. Obviously the hard the strategy the more intermediate ones will appear but for these mid range ideas I think I can get a good sample set together. I'll publish the links on the strategy pages.

I think I know what you mean. But it might be complex to realise. The solver has been designed more to show strategy steps with minor tinkering of the candidates. It sounds like what you are asking for is more in the line of a 'player' in which the user takes full control of the solving process but candidates are cleaned off when a big number is entered. You can simulate this by entering a number in the small board and pressing "take step" once which will update the candidates. But that will reset any personally edited candidates. Since the solver/player doesn’t know what are personally edited candidates it doesn’t respect them if they should be removed. Quite a difficult issue

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.