Unfortunately this is an inherent limitation of estimating workloads from a ramp test without physiological data to reveal what is actually happening in the body. The particular peak workload/power you land on will be protocol dependent, therefore so to will be any threshold estimate if it’s based on a flat percentage of peak power.
Shorter stages or higher steps will produce a higher peak power and higher threshold estimate. Longer stages or lower steps will produce a lower peak power & threshold.
Many (most?) studies use ramps of 20-30 W/min over continuous (ramp test) or 1-3min stages (incremental step test) and are designed to take the subject to exhaustion in ~12-15min. The specific protocol often depends on the target population, ie. more well-trained subjects will have a steeper ramp to reach exhaustion in 12-15min, while less-trained/general pop will have a shallower ramp to reach exhaustion in the same time.
So both Zwift & TR are in line with traditional ramp test protocol. I don’t think we can say either is overall more accurate than the other. It will depend on your individual physiology, which neither measure.
I’m less familiar with the design of the Zwift ramp test, but from the TR podcast and articles, I know they’ve dialed in their ramp protocol (6% steps, 75% of peak 1min power) based on a massive dataset from the TR user population. Probably the widest data set on ramp tests ever collected. So I trust the TR protocol is valid on average… but you don’t care about average. You care how accurate the test is for you.
IMHO, best thing you can do with the results of either ramp test is just start training knowing your FTP is probably somewhere within that ~15 W range. And dial in your workout targets as appropriate without being to beholden to a specific FTP number.