When I use that, the 55% is at year 4, and the model I am running "fits" the 133 and 208 event markers just fine. I think if I change the standard OAS to match 55% at 3 years then those markers will no longer match up. ( I will put it on the todo list )
So also if Geert says 55% at year 3 I don't know what that is so dramatic, because the article there suggests 55% at year 4 already to begin with. Maybe it has something to do with the special subset we are dealing with. Then there is the "observed" vs. "relative" distinction.
Feel free to complain how I do it, but see the "Slide OAS Setting" tab to see how the OAS % correspond to how many days into the trial.
What I am doing at the moment is not trying to curve fit formulae, but given the known characteristics, roll billions of dice against them.
I create a separation in the test and control groups by sliding their enrollment dates +5% and -5% to get the characteristics before rolling dice.
Then you can see the new first tab what that means.
In the monte carlo, the OAS% as of a given day means "can't kill any more patients today as of...this minimum of the enrollment for that person's enrollment month."
News 
Market Data 
Discover 
AI
