The part that gets me is testing with each breaker on. That won't do anything. The impedance of each bus bar is way lower than that of each circuit effectively shunting each circuit out.
The shiny object that is drawing all the attention is the tripping AFCI breaker. He starts off with the one AFCI not tripping upon test. . .installs another and it does not trip upon test when the branch circuit is connected, but it does test without branch circuit conductors connected to the AFCI.
Then he gets an arc weld arc when banging the disconnected branch circuit hot to the disconnected branch circuit neutral. A simple line to neutral solid short. That's it, right there.
The disconnected branch circuit, that is supposed to only be fed by the AFCI breaker, is still hot, energized by a connection to another branch circuit all together.
So, whether AFCI or standard breaker, both are going to try to energize a bolted short across L1 to L2 and blow with high fault current; or nothing will happen if they are connected L1 to L1, or L2 to L2. (But of course, the AFCI sensing circuitry will be backfed.) But this is Overcurrent protection operation. AFCI has no part in the underlying error.
So here, with the discovery that the disconnected branch circuit is still hot, MBrooke, connect the light between the disconnected hot and neutral (bulb lights up) and go down the circuit breakers turning them off, then back on, one at a time until the light goes off then on.
Then go inside, armed with the label info and hunt for the bad splice in a junction box on the two circuits.