I’ll assume that by using them in “series” you have one BR1 connecting (“BR1#1”) via ethernet (“WAN1a”) and cellular (“WAN1b”), and that WAN1a is connected to the LAN of the second BR1 (“BR1#2”), which (only) connects via its cellular WAN (“WAN2b”).
In that case BR1#1 can achieve load-balancing of its two WAN connections in the normal (“true?”) manner, with the outbound policies determining the particular load-balancing regime as per your preference. As pointed out by @erickufrin, this would be session-level balancing. As he suggests, you can define a SpeedFusion connection employing the two WAN connections of BR1#1, getting packet-level balancing (and other benefits). Still a “true” load-balancing of the two WANs of BR1#1.
In the above, BR1#2 is simply a very expensive modem.
Of course, you may employ other capabilities of BR1#2 (such as having other devices hanging off its LAN ports), making it a more worthwhile topology.
Things become a tad more interesting if BR1#2 employs both its cellular and its wired WAN options, in which case you get three connections from BR1#1 to the internet (WAN1b, WAN2a and WAN2b). You can balance across those three by working with the outbound policies of the two BR1s, e.g., combining a weighted balance of 2:1 on BR1#1 with a weighted balance of 1:1 on BR1#2.
However, if you connect BR1#1 with SpeedFusion employing WAN1a and WAN1b, then you will get the benefit of only one of the connections of BR1#2 (either a random choice, or the one determined by whatever outbound policy you have for BR1#2).
I thknk this is correct (wiser heads may correct me :-))
Cheers,
Z