The same is true for Barbs though except it is a bigger effect.This may be part of the reason some of us value shield differently.
If you are telling the player what the attack roll is, then your right, shield is very efficient.
If you are telling the player "the attack hit", then shield has a miss chance, and is toned down.
Statistically if all you know is a single roll succeeded then if you make a population estimate based on that single sample, the estimated population median roll is success. This means in a completely blind test the chance of another sucess on the reroll is higher than the chance of a failure on the reroll.
This means statistically if you have no idea of the roll or the target DC, the chance of failure is less than 50% if the only observation is a success.