We know that, for instance, the measles vaccine causes a severe allergic reaction in 1/1,000,000 cases.
Anaphylactic shock in particular is something that, if it's going to happen, it's going to probably happen within minutes of the vaccine being delivered.*
So yes, we have the ability to measure this one potentially accurately (although the fact that doctors are not required to report such events to any central agency means that you don't actually know whether the incidence is higher than that or by how much).
This is not true, however, of other potential complications, like long term immunological influences, or kidney interference 6 months down the line, or who knows what all. People could be dying routinely and not being statistically recorded simply because a time delay makes the cause/effect unclear, and it's still rare
enough that individual doctors don't notice ongoing patterns and thus don't file reports which then can't be found by larger studies easily (even just 1/10,000 would be quite rare enough to fly under the radar with a moderate time delay and no routine investigations or required reporting system). You cannot simply assume that all possible complications work on the same timescale as allergic reactions.
*Although also, even allergies can be delayed reaction. Often, your allergic reaction system has a tendency to lie dormant until you have built up half a dozen different allergens and then it starts failing. So you could have a situation where somebody has 4 allergen sources, and is somewhat allergic to shellfish, but adding that one wouldn't have done anything to them. but then they get a vaccine. The vaccine isn't enough to tip the scales either, but adds to the allergen load. Then a week later, they eat some shellfish and now we are up to 6 and the system is overwhelmed and they go into shock.
Nobody would report that as a vaccine reaction, but in that example, it essentially is just the same as one for practical purposes.
and if the optimal level is unknown then we should surely be pushing for it to be as high as possible to be on the safe side
But you. don't. know. it's. the. safe. side. As high as possible might just as easily be the dangerous side. Which is why this is not a useful default policy.
"The enemy is either on the left or the right. Civilians are on the other. We have no idea which is which. But just to be on the safe side, we should shoot left as often as possible" <-- this is equally logical as what you're suggesting the policy be.