Scott, thanks for following up. Looking now at your example I'm trying to figure what is happening. Thinking aloud here, so apologies is some of the following are already obvious to you.
If I get you correctly, the path you are speaking about is : https://plus.wikitree.com/function/WTPathCycle/PathCycle.htm?WikiTreeID1=Stark-3232&WikiTreeID2=Steward-2309
First thing : the new profile Larwood-186 is not identified in the WT+ report because it's not in the WT+ dump since it was created after. See at the end of the report : Date of Data: 7 Jan 2024. This profile is by default marked as "private/unlisted" in the report, with no data. It's "node 1" in the first table, and "node 0" in the cycle.
Next week, when the dump is updated, this profile will appear in clear, with its ID, name etc, since it's an open profile.
But : this node is already known by the Connection Finder when the function calls it, since the CF is updated almost in real time. Hence the computed antipodal distances should not change when WT+ is updated (barring other changes like new profiles and/or connections which would change the local network geometry)
So, why is this cycle not geodesic, but "almost"? All antipodal distances are either 8 or 9. Since it's a odd cycle, each node has two antipodes. For each node, at least one antipodal distance is 9, for some nodes, both are 9.
This is a very frequent result, actually the most frequent, and there is a general explanation I will try to add to the documentation at some point, and hopefully what I explain here is a kind of blueprint for it.
The distance found means there is a "chord" somewhere in the cycle, that is somewhere in the path a triple (a,b,c) where a and c are at distance 1. It cannot be in the path from 1 to 18, otherwise this path would not be a shortest path in the graph minus node 0. So it's either (0,1,2) or the other way round (0,18,17). Not (18,0,1) since we know that one is of length 2.
Checking manually or with the CF : Larwood-186 (0) is at distance 2 from Morris-1608 (2) so that is not the shortcut.
But Larwood-186 is at distance 1 from Larwood-74 (17). They are siblings, and that is the responsible shortcut.
Now we know which node to strike off the path to get a geodesic cycle, it's their common mother Steward-2309 (node 18). Checking the path from Stark-3232 to Larwood-74 ...
https://plus.wikitree.com/function/WTPathCycle/PathCycle.htm?WikiTreeID1=Stark-3232&WikiTreeID2=Larwood-74
... you get as expected a geodesic even cycle of length 18.
Quod erat demonstrandum
Note that I had to refresh the page 3 times in the first example to get all the antipodal distances because of the timeout of 30s, which is not a bug but a feature, as explained by Aleš somewhere in the other conversation.
The even cycle is processed twice as fast, because you have only 9 distances to check.