Last night I encountered the following situation at FTDNA:
My Mom matches a mother and a daughter. Let's call them Winnie (the mother) and Paulette (the daughter).
Mom, Winnie, and Paulette all match on the same segment of chromosome 21.
I compared Winnie to my grandmother (Nonna). They don't match, so I assumed (ha!) that this match is on my grandfather's side.
However, then I noticed Paulette among Nonna's matches. I assumed (ha HA!) that it was a connection on another line. So, I was very surprised to see that Paulette and Nonna match on that same segment of chromosome 21.
So! Mom, Winnie, and Paulette match on a segment. Mom, Nonna, and Paulette match on the same segment. Nonna does not match Winnie.
In all cases, the shared segment is 10 cM, just over 5 Mb, and over 1000 SNPs.
I'm guessing that FTDNA uses a 5 Mb cutoff like Ancestry does, and maybe through general flukiness, Winnie just missed that cutoff when her test was compared to Nonna's? (I'd ask this on the FTDNA forums, but I like it better here.)
Lesson learned: I'm going to be all the more cautious now about taking sides if the match is low in Mb or if I'm basing my assumptions on the absence of a match. If Winnie's daughter hadn't tested, I'd be running down the wrong trail entirely.
(Unless this is all IBS chaff and I shouldn't be running down any trails at all.)
P.S. Amusingly, I only match Winnie at FTDNA. I did consider this scenario:
Mom, Winnie, and I match on chr 21 on my grandfather's side.
Mom and Paulette match my grandmother at, eerily, the exact same stretch of chr 21 (presumably through Paulette's father).
However, I do match Paulette at GEDmatch (Winnie's not there), so I assume it's all just a matter of unlucky wobbles in the algorithm and not an incredibly odd coincidence. :)