I think the keyword here is "convention." What kind of manipulations does the audience conventionally accept? In other words, if you effect a specific choice at any moment in your picture chain, is it assumed that your audience will have some awareness of it to "reverse engineer" that manipulation and get the point you're trying to make without believing that they're seeing something that isn't there.

Example #1: Removing red-eyes. Most people know what red-eyes are, and some even know why they exist. When you are removing them, it is assumed that you are removing an unnecessary photographic artifact.

Example #2: Black and white photojournalism. We all know that the world is in color (sorry for excluding daltonians here). Therefore, when we see a b&w story, we can infer that there is no intent to present a reality that sticks photon by photon to the picture. We know what to take and what to leave out.

Example #3: Explicitely reconstructing a scene. If, unlike NBC, you would have told your viewers that you are using pyrotechnics to recreate the explosion of a gas tank by adding a mention to this effect, you allow people to see past the "forgery" to get to your point, without deceiving them. If you do not, then you are flaunting that convention, and therefore misleading your audience.

There are many more thorny examples. I'm sticking to simple ones to illustrate my point, but the idea is that what is acceptable manipulation is not stipulated by a code, but rather by a convention which is shared by an audience and picture-makers.