The AA program is very exact in it's definition, allowing AA based on a drug showing effect on a surragate that has not yet clearly shown to be of clinical benfit (but obviously is believed to be of benfit).
The exact question of "can a OS benifit with less supportative data qualify for AA" was actually asked of the FDA in it's public comments period on the AA process. The FDA said NO, period.
I don't dispute the logic of the argument you and a few others make, but it has no relavency to present FDA decision making.
It's also irrelavent to Prov, it clearly gets aproved despite somewhat skimpy supportative data.