Early research always had 1.0% in its confidence intervals, which is most likely the right IFR during the first phases.

The 0.1%-0.2% was just bad science, taking medians over countries with lagging statistics reports.

Where did you find 3.4%? Isn’t that an upper bound?