idle delta
Studies have shown that the biggest factor in if a LoRa packet is successfully the time on air.
So the likely hood of two smaller packets both succeeding should be better than one bigger packet.
but if you just split it into two piece it's likely that when a node reboards casts it that you'll receive a fragment you don't need
Erasure codes allow you to split a message into N symbols and you only need to receive M of the min order to recover the original message
once you have the original massage you can generate the other parts
So why don't we send M parts then the nodes the success receive it broadcast a different M subset of parts
there are 'fountain codes' which can generate a potentially limitless number of symbols.
I'm still looking into is fountain codes or some other error correction code with a fixed number or parts would be best. taking into account the effort needed to encode/decode the message and the likelihood of receiving I symbol you don't need (for a message you haven't decoded yet)
crude lake