Yes it is unrealible by design but in the last 8 month, since my gateway and some nodes are running, it happens much ofter that the TTN Backend has a problem (my monitoring alerts me) than a packet from a node is not received by the gateway. I can live with this sometimes instable backend, my use cases working anyway, but i guess in a industrial environment it is necessary that you need a stable running backend on that you can trust it is working.