Haven't used the rating feature much, but something's awry in the way I'm using the program. Every puzzle I ask SE to analyze comes back with a difficulty rating of 7.2. I reloaded the program and got the same result. It's as if this version (1.2.1) maxes out at 7.2. Easier puzzles seem to yield various ratings. Here's three somewhat difficult puzzles. I'm wondering what ratings others get when they ask SE to analyze them:

- Code: Select all
`007260001300007600000900003409500000010000030000008204200005000006100005100076800`

000080000001706800060502030047000310100000006085000720030901040009803200000050000

071000430800400002400005090000001900040902060003700000060500008200003009084000510

As far as the analysis itself is concerned, the solver seems to rely on what it calls "Forcing Chains." Maybe a forcing chain expert (Dan?) could verify that these are not forcing chains but discontinuous nice loops. In fact, it appears that every "forcing chain" the solver reveals is a weak discontinuous nice loop. By "weak" I mean it starts on a weak link rather than a strong, like this example:

- Code: Select all
`If we assume that R1C9 contains the value 9, it follows, through a Forcing Chain, that R1C9 cannot contain the value 9. Therefore we can conclude that R1C9 does not contain the value 9.`

Most human solvers (around here, anyway) seem to rely more heavily on chains that start with a strong link. I've never seen SE come up with a chain that begins, "If we assume that R1C9 does not contain the value 9, it follows..."