Thanks me22ca, some great points as usual, I've updated the video on Youtube now to include details about the settings I used, but I can't seem to edit the description on Steam, so this comment will have to do :-)

I used mostly default settings, but with some changes to improve accuracy, in particular I used:
Simulation accuracy: Very High
Reckless drivers Percentage: 2%
Individual driving Styles: On
Disable despawning: On
Enable Advanced Vehicle AI: On
Dynamic Lane selection: 30%

I actually did spend a bit of time investigating using the lane connectors in different ways, including partial-turbo and full turbo-style lane connections and I couldn't find a single combination that improved on the "Just-STOP-signs-and-no-lane-connectors" configuration!

I didn't try altering the speeds from the defaults, thinking about it, it might improve the results of the YIELD tests in particular, I'll take a look.

(Hmm, I guess it's hard to compare vanilla stop signs since you can't turn off despawning without TMPE...)

Amazing work! Great to get quantified results that the yield sign is essentially irrelevant -- have you posted this to the TMPE github?

A few other things I wonder about:
- This was using default lane arrows, not dedicated-single-exit-lane, like the roundabout quick setup does. (And which generally gives the traffic a turbo-style flow.) I always thought that helps; do you know if it does?
- Was simulation accuracy were you running?
- How do vanilla stop signs (with TMPE unsubscribed) compare to TMPE stop signs?
- In real life the speed on roundabouts is much lower than the 100 kph of the highway road you're using there. Can one get more vpm with faster or slower roads?