From a comment by “Eggplant” at Belmont Club:
Supposedly the US has war gamed this thing and the prospects look poor. A war game is only as good as the assumptions programmed into it. Can the war game be programmed to consider the possibility that a single Iranian leader has access to an ex-Soviet nuke and is crazed enough to use it?
Of course the answer is “No Way”.
A valid war game would be a Monte Carlo simulation that considered a range of possible scenarios. However the tails of that Gaussian distribution would offer extremely frightening scenarios. The Israelis are in the situation where truly catastrophic scenarios have tiny probability but the expectation value [consequence times probability] is still horrific. However “fortune favors the brave”. Also being the driver of events is almost always better than passively waiting and hoping for a miracle. That last argument means the Israelis will launch an attack and probably before the American election.
These are important points. The outcomes of simulations, including the results of focus groups used in business and political marketing, may be path-dependent. If they are the results of any one simulation may be misleading and it may be tempting to game the starting assumptions in order to nudge the output in the direction you want. It is much better if you can run many simulations using a wide range of inputs. Then you can say something like: We ran 100 simulations using the parameter ranges specified below and found that the results converged on X in 83 percent of the cases. Or: We ran 100 simulations and found no clear pattern in the results as long as Parameter Y was in the range 20-80. And by the way, here are the data. We don’t know the structure of the leaked US simulation of an Israeli attack on Iran and its aftermath.
It’s also true, as Eggplant points out, that the Israelis have to consider outlier possibilities that may be highly unlikely but would be catastrophic if they came to pass. These are possibilities that might show up only a few times or not at all in the output of a hypothetical 100-run Monte Carlo simulation. But such possibilities must still be taken into account because 1) they are theoretically possible and sufficiently bad that they cannot be allowed to happen under any circumstances and 2) the simulation-based probabilities may be inaccurate due to errors in assumptions.