LoRa is only 25–100 mW. Given that you mention 868 MHz you are probably using the SDR860 European standard, which means a maximum of 25 mW ERP.
Even a grossly mismatched antenna system wouldn't be a big deal in terms of safety (for people or equipment), at least for testing. Would it be wasteful using the energy to warm the antenna system slightly? Probably, but you won't care unless you are planning on sipping battery power for weeks or months. I suppose it depends on the device doing the sending, and whether it can handle the mismatch and potential higher voltages. But most LoRa equipment I've seen is made for more rugged IoT situations so I imagine the antenna connection will not be so delicate. (But you might want to investigate that anyway.)
So, to answer your question, you are probably ok to just experiment with the antennas you have, though the stubby will probably be the most efficient -- in free space with a decent ground of some kind.
By my back-of-the-envelope calculations a $1/4\lambda$ for 868MHz would be approximately 8.6cm. This might mean you are feeding the antenna at location with a higher voltage node. Well, maybe. We don't really know anything about this antenna, other than its length. And who knows what the ground plane looks like to the monopole.
But who knows? Many commercial LoRa external antennas are $1/2\lambda$ in length, so basically a vertical dipole with whatever counterpoise/ground it can find.
You may have stumbled upon a reasonable configuration because the always imperfect ground offered by the car body and the extra length of the monopole combined to give you an accidental lower feedpoint impedance and a lower takeoff angle resulting in decent results.
Honestly, there is no way to really know without analyzing the system in place. Otherwise, experimental results speak for themselves.
Not to mention that the whole point of LoRa is to provide bursty, low-bandwidth, low-power "connected" networking comms in marginal situations. So the mode is rather forgiving. If you can get a signal out enough to "connect" then that's what matters.
Since you are experimenting anyway, why not simply hook up piece of large gauge single-core copper wire cut to the approximate correct electrical length and then trim until you get the result you want from 2-3 listening nodes?