I don’t think that’s entirely accurate, and also doesn’t explain why the effectiveness of LTC on output 3 should be a function of the level on outputs 1 & 2. The grandMA appears to expect an analogue signal between -11dBu and +15dBu:
https://help.malighting.com/grandMA3/2.0/HTML/fs_connect_smpte_ltc.html. How that relates to “0dB” (presumably you mean 0dBFS?) depends on the gain structure of the audio interface.
In fact, since LTC is a continuous audio signal, you often find you have to drop its level to avoid crosstalk in analogue multicores. Perhaps that’s the reverse issue here: crosstalk from the audio signal is disrupting the LTC signal post interface? The Focusrite 18i20 has a maximum output level of +16dBu (which would clip the grandMA’s inputs). Adjusting the levels so that the LTC signal is leaving QLab at -16dBFS should generate a healthy 0dBu analogue signal, in the middle of the grandMA’s acceptable range.
It could also be something really stupid like the Focusrite 18i20 not having enough power and therefore dropping signal levels once the load becomes too high… Does something exciting happen in the audio at 5s that would correlate with a dip in LTC level? Is there a parallel box available to split the signal and see what is going on with it – maybe by routing it back into an input to monitor?
Rich