[ this thread is a few months old, but I don’t think the topic is stale and I’m wondering whether my interpretation is plausible]
I’m wondering whether the answer to the original question “why don’t I see a gradual drop-off of signal strength” has to do with the noise level. In the 2nd mapper screen shot posted by Jeff-UK it shows an SNR of -5.75dB and a signal of -103.75dB. The demodulator is spec’d at -7.5dB for SF7, so that’s pretty close, there’s very little headroom left to go further.
In contrast, the prior screen shot posted by Jeff-UK shows a signal of -123.5dBm and SNR of -8.5dB (it’s SF8). But the big difference here is an RSSI, i.e. noise floor, of -115dBm while the OP’s test has a noise floor of -98dB. So the OP was looking at the absolute signal strength instead of SNR. The OP’s tests were at the range limit even though the absolute signal strength wasn’t “that low” due to the noise floor.
Indeed both RSSI & SNR have to be considered. The examples I called out above have, by definition, gone through and its always a challenge to determine those that didnt go through and the circumstances (wrt RSSI, SNR, Reflections, Absorbers etc. that prevailed at that specific instance…noise is rarely constant and at any instance will vary by underlying noise floor and any incremental overlays.
The challenge when trying to judge if a link is possibly ‘too’ close to the SNR limit for comfort is to try and understand the prevelance of e.g. impulse noise inputs or any ‘bursty’ noise sources that may come and go and perturb a signals resiliance. The fact is that LoRa modulation also behaves differently in the presence of different types of RF interferers…other LoRa signals, the SF’s of such, legacy signals - gFSK, FSK, OOK, FM, AM etc, as well as other more complex modulations (close proximity out of band cellular? etc.). LoRa is quite resiliant c/w legacy modulations and many very narrow band signals, plus it has the delightful property of even being able to reconstruct a message where up to half (IIRC) of a given symbol is lost due to interference or simply not heard! It is therefore a good performer under near/far problems and has good signal coherence wrt motion induced doppler etc. (Latter clearly SF related, which can be a issue with e.g. trackers/asset locators/mobile condition monitors etc.) That said either a symbol is detected or not leading to brickwall effects so where as the human ear can detect a message transmitted over say a noisy FM or AM signal like all digital reception mechanisms a signal can be recovered until suddenly…it cant! At that point error concealment or error recovery mechanisms might cut in to save the day, but even FEC as used for LoRa/LoRaWAN has its recovery limits.
RF is often described as a “black art” for a reason and IMHO LoRa/LoRaWAN is borderline wizardry or white witchcraft which is in part why I have advocated it for a dozen years now! (1 month now to formal 12th Anniversary of the public announcement of the technology! )
At that point error concealment or error recovery mechanisms might cut in to save the day, but even FEC as used for LoRa/LoRaWAN has its recovery limits.
(1 month now to formal 12th Anniversary of the public announcement of the technology! 
)