This article indicates that the USA FDA allows for up to a 20% margin of error for values required to be on the Nutrition Facts label. The article also describes multiple methods of measuring the calorific value of food, either by burning it in the not-so-TSA-friendly-named bomb calorimeter or through methods that have standardized the calorific value of each constituent nutrient.
It’s also the case that not every foodstuff is perfectly identical to all other products. A banana is hardly going to be constituted exactly like another banana, and even the most basic measurement of mass will not match up to other bananas. Yet some sort of “standard” banana must be assumed in order to print the nutrition label.
As an aside, I do fondly remember making a bomb calorimeter in chemistry class using a polystyrene cup as the insulation. It worked remarkably well, IIRC, being within 10% of what we were given as the expected value. Obviously, real measurements would be far more controlled than what some college freshmen can manage, but the concept is sound, if only measuring what the food provides, not necessarily what the human digestive tract can extract.
As an aside to an aside, celery appears to , even after considering human digestion.