How accurate are the calculators

There are a number of formulas out there, which is one reason why you see a difference in caloric recommendations.  Another reason you will see a big difference is due to how these calculators determine your activity level, which can be all over the map.  If you are interested in a method that uses your lean body mass percentage, try the calculator here (use the advanced options to see the different formulas and enter your bf%).  Obviously, this method assumes that your body fat percentage is indeed accurate, which is a whole 'nother can of worms. 

There are other factors that make all of this even more confusing... For instance, if you estimate your daily caloric needs with no activity and then add in the total number of calories burned from an hour on the treadmill, you're going to be overstating your calories burned (because you've already counted the calories you would have burned just existing for that hour).  It's SO easy to overthink this.  Your best bet is to pick one calculator's method, then accurately log all your activity and food intake.  Stick with it for two weeks and then see where your weight ends up.  Adjust accordingly and revisit a week later.  If you're consistent with how you log, and you don't have any extenuating circumstances, then you'll end up with a fairly accurate calorie target for weight loss (remember to readjust as your weight drops).

I'm not sure why you think you don't use fat during exercise?  Sure, it won't help you lift a barbell, but it definitely does get used as fuel.  The number of additional calories burned from lean mass is really not that much (seriously it's around 5-6 calories per pound per day).  You generally won't be putting on muscle if you're holding a caloric deficit, so don't worry about it.  The calculators give you a number to maintain your current mass, which is how a good intake for weight loss is determined.  It's also how your calories burned are estimated.  This is why you should include your total weight in the calculators (or just use the one I linked with bf%).