.
As explained in previous posts, I have
a systematic procedure for generating extreme B T&E(1) puzzles from minimal T&E(3) ones, or from their min-expands. As previously, I consider "extreme B" to mean "B ≥ 12" (justified by the unbiased B distribution).
Reminder: the procedure is very simple (but time consuming in the "compute minimals" part): take the 1-expands of the T&E(3) min-expands, filter those in T&E(1) and with B ≥ 12, compute the minimals of the latter and filter those in T&E(1) (which are guaranteed to have B ≥ 12). Notice that, for each puzzle in T&E(3), the whole process remains within the same solution grid (in particular, there's no {-p +q} search.
As mentioned in previous posts,
the mean throughput is 1.28 B12+ puzzle for each T&E(3) min-expand. As far as I know, this is the first time some kind of extreme puzzle can be obtained from another kind in a systematic way. What's also noticeable is, no similar procedure works to obtain high BxB puzzles in T&E(2).
I've just published on my Google drive a zipped collection of 504,481 such puzzles with their B ratings (corresponding to only a part of mith's T&E(3) collection):
https://drive.google.com/file/d/1wSkSokvR8maE9yQJI3_X1NSb5ank35IU/view?usp=share_linkThe distribution of B ratings in % (in the [12 24] interval is as follows:
- Code: Select all
12 13 14 15 16 17 18 19 20 21 22 23 24
45.02 25.81 14.09 8.88 3.70 1.54 0.62 0.23 0.091 0.018 0.0040 0.00079 0.00040
For the most extreme values, this means
- Code: Select all
3126 B18 puzzles
1141 B19 puzzles
461 B20 puzzles
92 B21 puzzles
25 B22 puzzles
4 B23 puzzles
2 B24 puzzles
.
