Graphs of DNSWL performance according to SpamAssassin Rule QA results.
Last update: 20190914.

On 2011-05-05, for bug 6557, the maximimum age limit for emails was changed to match what is used for score regeneration, from:

my $OLDEST_HAM_WEEKS    = 52 * 10;      # 10 years
my $OLDEST_SPAM_WEEKS    = 6 * 4;       # 6 months
to:
my $OLDEST_HAM_WEEKS    = 72 * 4;       # 72 months = 6 years
my $OLDEST_SPAM_WEEKS    = 2 * 4;       # 2 months

Percentage of emails per trust level hit, which are not spam:
Current values with some possible short term goals:
99.9078343398791% of email DNSWL ranks as "high" is non-spam. (Goal: 99.99%)
96.4883128019604% of email DNSWL ranks as "medium" is non-spam. (Goal: 98%)
85.3648228400161% of email DNSWL ranks as "low" is non-spam. (Goal: 85%)
81.3109693839974% of email DNSWL ranks as "none" is non-spam. (Goal: 50%)

Percentage of ham that hits DNSWL per minimum rank:
Currently:
1.2907% of non-spam comes from an IP with a rank of "hi". (goal: 12.5%)
2.5557% of non-spam comes from an IP with a rank of at least "med". (goal: 25%)
4.3591% of non-spam comes from an IP with a rank of at least "low". (goal: 50%)
17.0958% of non-spam comes from an IP that's listed in DNSWL at all. (goal: 100%)

S/O, SA's usefulness score per trust level:
0 = 100% non-spam (best)
0.5 = useless (half spam, half non-spam)
1 = 100% spam (worst for non-spam indicating tests like DNSWL)
This is basically a percentage of spam, with the numbers of spams and non-spams normalized.

Rank, 1 is best, 0 is worst:
I think they sort all SA rules by S/O (above) and scale the rank numbers to the range of 1 to 0.

Total count of spams and non spams (not related to DNSWL)
Current non-spams: 345467
Current spams: 1371135

Total percentage of email that was not spam (not related to DNSWL)
Current percent not spam: 20.1250493707918