Update – CQWW 2023
I haven't done this in a couple of years, but herewith an attempt at an update on the RBN, following CQWW CW.
Our online stats show that we had over 250 nodes online during the contest. A fair number of those were only online for the contest; right now, we have 207, which seems to be about normal over the last couple of years. All told, 10,881,462 spots were posted during the contest, almost exactly the same as in 2022, an average of about 227,000 per hour, or about 63 per second.
One topic that has become near and dear to my heart is reducing busted spots, even though the RBN averages close to 99 percent right. I got particularly interested because I was being busted often as TN4ZR, particularly during CWTs. With some experimentation, I found what seems like a solution for those (like me) using N1MM Logger or other stored CW software. All I needed to do was insert a half-space character (~) between the word CWT and my call, so instead of sending CWT N4ZR, I was sending CWT ~N4ZR. The added delay is barely perceptible (to me), but it seems entirely to solve the problem - I ran for several hours during last weekend's CQWW, was spotted hundreds of times, with nary a "TN" in sight.
Limited data from CQWW suggest, though, that the extra half-space character may be less needed if what comes before the call is "TEST", which is one of the keywords CW Skimmer uses to identify a transmission as a "CQ" and therefore to be spotted. One big multi-op was spotted about 25,000 times each day and only ran into the "leading T" error about 0.6 percent of the time, while the problem in CWT, for me and others, had been as much as 2 percent. Ultimately, this suggests to me that the added half-space may be a good idea, regardless of where the "T" comes from.
VE3NEA is quite protective of his intellectual property in the CW Skimmer software, but from what he says I'm not surprised that small changes like this can have a big impact, particularly for "run-together" busts. He says his code "recognizes" some combinations of letters and therefore "resists" running them together with callsigns. TEST is one of the recognized combinations, which may be why "TEST N4ZR" is busted less often. Busts like N3AD/N3AM will probably never be totally soluble.
Sidebar: Improving the number of busted spots you see, but what about the spots you receive? There are a couple of ways to weed out many of the busts.
VE7CC's cluster client software does this as a matter of course, and he explains that in detail on his website, at http://www.bcdxc.org/ve7cc/. In addition, his node software incorporates a lot of filtering in an attempt to catch as many busted spots as possible before they are even distributed to "retail" nodes.
Many users of AR Cluster V6 use CT1BOH's filters to improve received spot accuracy. The filters Skimvalid and Skimbusted, added to your instructions to the ARCluster V6 node you use, can be very helpful. Skimvalid looks back to see if there are two or more spots of the same call, on the same frequency (+/- 0.2 KHz) before marking the spot as "Valid". Obviously, that will help with single busts. Skimbusted works by looking back to see if there is a similar but different spot, already determined to be Valid and posted on the same frequency (+/- 0.1 KHz). A similar call is one that can be transformed into the new call by character insertion, deletion, or substitution.
A full explanation of all the AR Cluster commands is available online at http://www.k3lr.com/w9zrx/AR-Cluster%20Filter%20Commands.pdf.