@incognico: Excellent post. Glad you have the gumption. I definitely look forward to more. This is definitely an area where even a single person can have a great impact on expanding the collective understanding.
For this type of LoRa listen mode to be attractive, a key question will be how long it takes for the CAD. With that information, and the current drain incurred during CAD (is it the same as during regular Rx?), and the current drain while sleeping (from the datasheet), one can start to estimate the average current drain, and hence battery life.
For comparison, on the RFM69, at 200Kbps, without much effort one can receive a packet with a listen window of less than 1ms or, alternately, detect a high RSSI (which could denote "an intent to transmit" in Joe's approach to Listen Mode) in even less time than that. The guiding principle is, of course, to spend the least amount of time drawing high currents and the most time possible in deep sleep, drawing nano currents.
For those reasons, I'm guessing you'll probably want to do at least your initial LoRa experimentation using the highest speeds that the LoRa will allow, which, IIRC, will be around 37Kbps.
There has been a significant collective learning curve on how to use the RFM69 efficiently (with more ways still being uncovered/discovered/developed). Partly that's because the available documentation was mostly limited to a fairly dense, sometimes sketchy, datasheet that in many instances has required a lot of unpacking and interpretation and experimentation just to figure out what it even means. I think there will likely be some kind of learning curve for the LoRa aspects of the RFM95, but straight out of the box the RFM95 does seem to come with more app notes and a somewhat better calculator.