I don’t believe it’s possible to devise rigorous rules that govern when lethal force can be used—the potential situations an officer can face are too varied. Attempts to come up with simple criteria, such as “imminent risk of bodily harm,” are impossibly vague, because the risk of situations can’t be quantified.
An officer driving his car on the freeway is in imminent risk of bodily harm, because there’s a probability that he could be in a crash at any moment (traffic accidents kill more officers then shootings). An officer raiding a meth lab is also in imminent risk of bodily harm. The difference between the two is the degree of risk, not the presence or absence of it. Any risk-based guideline is necessarily subjective.
Because risk is a continuum, I think it’s wrong to rely on absolute statements when analyzing any particular type of case. In the case of an officer being threatened with something that resembles a weapon, many people have made statements to the effect that lethal force is justified unless there’s certainty that the weapon is harmless. Should officers be allowed to shoot down theater actors who point prop guns at the audience? There’s no way to be certain that it isn’t deadly—actors have been killed by supposedly empty guns. This is obviously a ridiculous example, but that’s the point: it’s a ridiculous standard. Even people who claim that certainty is the correct standard to use are probably really using a more nuanced standard that makes room for minimal levels of risk.
When looking at certain types of situations, I think the degree of risk needs to be analyzed at all levels. Suicide by cop seems an appropriate situation to talk about, for some reason. 
Many people have stated that a person who is suicidal is mentally unstable, and more likely to harm the officer. I disagree with this statement. Yes, someone committing suicide by cop is mentally unstable. But I don’t think it follows that the person is homicidal—in fact, in the Florida case, the victim was obviously not homicidal. What’s the probability that the average person who commits suicide-by-cop is homicidal (excluding cases where it’s already known they’ve killed someone)? I don’t know, but I think it’s an important statistic.
People have also been using absolute standards to judge the utility of less-then-lethal weapons. Again, I think that this is inappropriate. Yes, there are people who will be capable of returning deadly fire after being hit with a bean bag or pepper ball. But that’s not the case with all people, and I’d be surprised if it’s the case with the majority of people. What the probability that a person could effectively return fire after being hit with a beanbag? I don’t know, but I think it’s an important statistic.
In order for an officer’s death to be the outcome of a suicide-by-cop “man points weapon at cop” incident, there are many different probabilities that come in to play. I think some of the more significant ones are: the probability that the person is using a real gun, the probability that the person has homicidal intent, the probability that less-then-lethal weapons will fail to sufficiently incapacitate the person, and the probability that the person will hit the cop with his return fire before he himself is killed by the lethal weapons of the officers present.
As I’ve said, I don’t know what those numbers are–if anyone has some hard, impartial stats I’d be very interested in hearing them. I’m inclined to think that they’re actually fairly low. I don’t think that most suicidal people are homicidal. I think less-then-lethal weapons will incapacitate most people. I think that shooting someone with a gun is a tougher skill then most people think, and not necessarily one that a mentally distraught person will be good at. The probability that the gun is real is going to vary greatly from incident to incident, and is one that the police will have the least confidence in. But it’s one that certain events can increase their confidence in (being told by a relative that the gun is fake would increase it a little, being told by an independent observer that the gun is fake should increase it a lot).
Given those beliefs, the probability that an officer would die in a suicide-by-cop incident is actually fairly low: taking the probability of the weapon being real as 90% and the rest to be 40%, the probability that officer dies is about %6. It’s a crude calculation, but it demonstrates how the outcome of a sequence of events can be far lower then the probability of any single event.
Given a hundred such incidents, there will probably be at least a hundred deaths (the suicidal person, and any officers that also die in the incidents) if the officers must shoot the suspect as soon as he points the gun. If the default behavior is to use a less-then-lethal weapon, there will probably be at least six deaths, although more of the total deaths will be cops.
Is this a fair trade-off? I think so. If you think that sounds callous, and that no probability of officer death is acceptable, consider that we send officers into dangerous situations all the time to accomplish goals far less noble then minimizing the loss of human life. If this sounds like cowardly Monday-morning quarterbacking to you, then that genuinely frightens me: The actions of our law enforcement officers, and the policies that shape them, must ALWAYS be open for debate and discussion. They are our servants, and we don’t want to let that tail start to wag us.