It seems like the most "scandalous" portion is just being
taken out of context.
Lets say I try to model vehicle acceleration, and I have a component when power transfer is consistent, and when it's erratic due to slippage (clutch, tires, etc). Now, I can record the intervals for the whole acceleration event, and based on that I realize that my model for normal acceleration is pretty good, but that my model for the portion when the clutch and tires are slipping sucks, it doesn't match what's actually happening. So I go get actual data on the period where the wheels/clutch are slipping and use that instead of my crappy model that isn't accurate. Does that mean that I'm falsifying data to cover something up? Hell naw. The model I created had a crappy portion , so I just replaced that part w/ real data, getting a more realistic model than what I originally had. Ideally, I'd love to have something the could explain the whole event, but if a portion of my model sucks it's much better to replace it w/ concrete data, even if it's kludgy, instead of using a flawed model.
In terms of the funding, I've had college faculty say that they altered the abstract for their research so that it would appeal more to homeland security and get some of that dough. It's common practice in research to create a proposal that has more appeal so it's more likely to get funded, just like it's common to shape a resume so that it appeals to a specific employer instead of just using the same one over and over.