Having looked at Keen's clue selection code, I also notice that the

intended behaviour of classifying multiplication clues as low-quality
if they only left one possible pair of multiplicands has never
actually worked, because I should have compared the possible clue
count against 2 rather than 1 since the multiplicands can occur either
way round.

[originally from svn r9165]

diff --git a/keen.c b/keen.c
index 175db55c323222d6a202c21c088f48a0b580d1b9..8d9b6bd8f493a8c8fb33f94d2b2fa92310fe4f71 100644 (file)
--- a/keen.c
+++ b/keen.c
@@ -931,7 +931,7 @@ done
                for (k = 1; k <= w; k++)
                    if (v % k == 0 && v / k <= w && v / k != k)
                        n++;
-               if (n > 1)
+               if (n > 2)
                    singletons[j] |= F_MUL;
                else
                    singletons[j] |= F_MUL_BAD;