(1) Arguments from the present. We could say that science knows of no shape shifters today, for example.
Though I doubt Bob will respond its worth responding to point 1 in greater detail.
The argument made by Bob, and the arguement made by most skeptics is that science somehow "knows" that there are no occurrences of X in the present, and as such, we believe that with high probability stories of X in the past are false.
First off, its worth noting the phrase "science knows". Science does not "know" anything about specific ocurrences of events. Science can tell us if a certain event is in violation with established scientific "laws" or if an event is not in violation with those laws. That is it.
Most events however, given specific descriptions, are not in violation of scientific laws. Let's take shapechanging for example. Let's take alchemy for example. Specifically turning lead into gold. Is this impossible?
The difference between lead and gold is the differences between the number of protons and neutrons in the different atoms, as well as the arrangment of the corresponding numbers of electrons. To turn lead into gold we must take away the correct number of each particle (specifically removing 3 electrons / protons and 8 neutrons) from lead.
Now, science can tell us that certain methodologies are incapable of turning lead into gold. For instance, it can tell us that chemical methods cannot turn lead into gold as chemical methods cannot strip away elements from the nucleus. But if all I know is that "someone turned lead into gold" science cannot say that this is impossible. Moreover, science cannot even say that it is unlikely. For science to say that the event is "unlikely" is for science to make a probabilistic statement. But by what methods could science make such a statement? One possible way is that the atheist could say that since many methods to turn lead into gold do not work, there exists no working method. This is fallacious inductive /statistical reasoning however. By conducting a sampling of the space of all possible alchemic methods, (if the sampling is a good one) we may conclude only that there are very few methods by which lead can be turned into gold at best. This is a conclusion that would not be debated by the alchemist at all. I would also say that we have no way to conduct a good sampling of this space, as the space of all such possible methods is essentially infinite. From an inductive reasoning standpoint the idea that "All methods I have thought up do not turn lead into gold, therefore no such method exists" is very weak induction.
So at best we find that when science is used to judge a pass event it can at best, say that the event is in violation of "scientific laws". Moreover, science can only due this when the method behind acheiving the effect has been stated. If we only know that an effect has occurred, science cannot say or "know" whether or not the event has happened at all.
Now there is a different type of argument to be made. Namely, we disregard science for a bit, and we no longer consider whether or not event X violates scientific laws at all. Rather, one says that event X is "improbable" and therefore given a past description of event X, there is a more likely explanation than "event X actually occurred". I'll deal with this later.