Thanks to those who provided helpful information, I should have some options here to start from a known-good solution and work from there. I’ve shut down the old equipment for now and started with a completely new device platform with (so far) better results.
For others who may have the same problem, also refer to the helpful feedback I’ve gotten in this (somewhat related) thread. While I don’t have a complete solution, it appears that MAC v1.0.3 is more reliable than v1.0.2, despite the devices working well for me on V2 that way.
I avoided asking a very specific question about one single combination of gateway and end-node hardware precisely because the answer to my general question would help restore my faith in V3 in general without debugging a specific set of hardware. Based on the feedback here, green-field deployment of V3 should work without any problems, so I’ll revisit the specifics and if needed, ask specific questions about the particular gateway and end nodes I’m developing with.
I’m considering this issue closed for now, but I am curious about the following, if any of you care to elaborate:
@kersing: Thank you - I will try some of the device brands you mentioned. Can you confirm that when you successfully added new devices to V3 that there was absolutely no communication with a V2 gateway, which in my case allowed devices to join, and then forward through V3?
@descartes: (Same question: Can you confirm that when you successfully added new devices to V3 that there was absolutely no communication with a V2 gateway, which in my case allowed devices to join, and then forward through V3?) but what hardware have you used (you didn’t list any)? In green-field deployments, without a V2 gateway to fall back on, the answer doesn’t seem quite as clear.
@Jeff-UK: I appreciate the graph of connected gateways, but being connected versus actually allowing a device to join V3 without the presence of V2 infrastructure are a little bit different questions.
All: you might understand the skepticism from those of us in countries with few or no V2 gateways in range of our devices. In high-gateway countries, it’s likely that anyone testing a new V3 deployment would also be in range of a V2 gateway, therefore hiding this type of problem (if it ever existed).