The Iridium 7 mission, Jul 25, 2018 gives a partial practical answer.
Throughout the webcast John Innsbruker reminded everyone that conditions for landing in the Pacific for the JRTI (Just Read The Instructions) ASDS were the worst they have ever tried to land in.
Yet they continued the launch when they had a back up launch window 23:55 hours later.
This was the first flight of that Block 5 core, of which only 4 are in the fleet at the moment. (1046 is being disassembled after first flight. 1047 at the time was on OCISLY coming back to port, this mission was 1048, and 1049 is being prepped for a mission in 12 days). Their remaining manifest for 2018, in the short term, is dependent on recovery and reflight of every stage.
Core 1046 is out of service indefinitely. They have 16 or so more flights to do, and the factory cannot make cores that fast.
Thus in some ways, this is the literal worst case. At a moment when there are not enough cores in the fleet to absorb a lost core without affecting the manifest (I.e. the part of the buisness that makes all the money, actual launches) and on the first flight (most expensive moment in a flight), they still launched when they had a backup window only day later.
Now it is possible that even though these were the 'worst' conditions they have ever tried to land in, it was still good enough, per their models to actually land.
But this is a pretty good data point on how much risk they are willing to take.