Robo Gate
CleanFeed.org is a use of advisory, third-party cancellation messages. groups.google.com ignores them. Berlin Free University heeds them. Such cancellations apply against the most persistent and obvious commercial postings. After 1994, some people got so adept at writing killfile methods for their own personal use, that they turned around and volunteered to offer the same service for other people. Jeneral Note: If a bot cancels your message, then you are being both predictable and annoying in a group full of people who do not care about MI5 persecutions. It might even say so in their posting guidelines. As a rule, if a post goes to more than five groups, then it is excessive crossposting (ECP), and posting the same message in several groups (MCP) is about the same problem, which a bot deals with. It is not impossible to make a post about international politics relevant to more groups than five. The difficulty in finding all those groups is not worth the trouble, and you might not want to jenerate heat outside of your own nation, anyway.
Flap Gate
This style of moderation is for carefully constructed traffic on a list that is effectively closed to anything but the most thoroughly researched documents. If one moderator raises an objection, then it does not get posted. Newsletters are structured this way, so that all feedback goes through fingers of at least two people, one of which has decisive access to a mailing list. It is prone to a flaw in a trap gate if has only one address published for feedback or contributions. I do not mean to underestimate how much power should be given to one voice in ignoring other voices.
Any Gate
This style of moderation is expensive in terms of demands on both moderators and list readers. All moderators see all input, and any one of them may let it pass without modification. If a message does not pass through such a process, then it can be presumed to offend stated aims of a list: such a message is unsolicited commercial e-mail, not conformant with the name of a list, or non-conformant with guidelines stated for a list.
Trap Gate
This style of moderation assigns moderated members of a list to a single moderator, who should provide feedback if a post is unacceptable. I do not recommend this method, because it can raise questions about style, personality, and level of throughput in a moderator if feedback is too terse or not reasoned in consensus of guidelines. It is even harder for a list to maintain an image of reasonable transparency if a moderator can assign himself to a list member.
Random Gate
This method is best for lists with high levels of traffic and a principle that includes openess. Moderators assigned to users are random for every post, so that load on a moderator is divided by a number of moderators, and little room is in it for the possibility that a moderator could block any particular user for personal reasons. It is in keeping with the idea that trolls are not people, but posts that do not conform to posting guidelines.
BrewJay's Noyz Jenerator
Following this is code for the pseudo-random number jenerator that wrote my background. The easiest way to make a pseudo-random number jenerator that does not produce anything like a flat histogram (or a pseudo-gaussian histogram in three-colour mode, looking at the grayscale) is to make it too complicated. The Lempel-Ziv-Welch compression method employed in GIFs and pkzip cannot compress my background at all (even PNG got slight expansion), and if I load it into Paint Shop Pro as a three colour imaje, then it has a pseudo-gaussian grayscale curve, which /dev/noise under Linux does not hav. /dev/noise under Linux has a level cut in the middle on a histogram in RGB mode. I made Paint Shop Pro read output from the following code as a two-fifty-six colour grayscale in raw mode at six-seventy-two by 930 (printer width and height for Internet Explorer with one inch marjinz all around), then dropped levels of red and green in it to zero:
OPTION BASE 0
DIM prand AS LONG
DIM seed AS LONG
DIM offset AS LONG
PRINT
OPEN "noyz" FOR OUTPUT AS #1
' 9*672/6.5=930
' Dimensions: 672*930
FOR seed = 1 TO 624960
' The starting point, one, is a seed, and it would be passed by reference, along
' with a length, so that the loop ends at seed+length, normally, and my purpose
' really is to make noise with it, so I do not really need an arbitrary precision
' math library. I will make a two-colour print at 1200 dpi, just to be sure
' that it is as good as I think -- and to get a good perspective on how small
' dots are at that resolution: 1200^2*10.5 = About fifteen megabytes (that does
' not compress) on one page -- hmmm. ran into a file size limit at the library.
100 prand = (offset + prand XOR seed) * 509 MOD 257
IF prand > 255 THEN
offset = offset + 1
GOTO 100
END IF
' Four lines above prevent zero from occurring twice as frequently as other numbers,
' and from 256 raising an exception, below.
PRINT #1, CHR$(prand);
200 NEXT seed
CLOSE #1
END
That is histograms of data from my PRNG loaded as a planar RGB graphic. The red, green, and blue ones are nearly flat. The hardest one to explain is the grayscale. It is pseudo-gaussian, because it comes from averaging red, green, and blue. It has a bump at the high end, because yellow is closer to white than blue is to black. Luminance, a cousin of the grayscale, is much neater in yellow, because it is a mean. Hue (in cyan) is almost flat. I do not know why the majenta curve (saturation) slopes upward, because colours naturally become more pure as they approach black. One property of my noyz jenerator is very hard to prove: it is not periodic. I can see that by inspection of my graphics. I see no eloquent way to prove it. To write more precisely, the period of my noyz jenerator is proportional to k, so that if k (or the range of values for k) contains one-hundred twenty-eight bits (or thereabouts), then calculating all digits of the period will take more time than Earth's projected lifespan. I hav not ventured into proving an exponential period in proportion to bits of my loop variable: It looks like that.
It is good, and not great, az gnu privacy guard 1.4.11 (gpg --no-armor --gen-random 0 144000 >graphic.raw) gets about four times az many colours in the same process. Although it haz about the same distribution curves in a histogram (1.4meg), those curves (and lines) are smoother than mine.
BrewJay's Babble Bin
