The water effect
Rice and pasta absorb water, so a cooked gram contains more water and usually fewer calories than a raw gram. Meat, fish, and many vegetables lose water during cooking, so nutrients can appear more concentrated per cooked gram. The batch may also lose fat as drippings or gain fat from oil.
This is why 100 g raw and 100 g cooked are not interchangeable database entries. They describe different physical states.
Match the entry to the state you weighed
If you weigh chicken raw, use a raw chicken entry. If you weigh it after roasting, use an entry that describes roasted or cooked chicken with a similar cut and skin status. For rice, match dry weight with an uncooked entry or cooked weight with a cooked entry.
The database description matters as much as the number. Look for preparation, drained status, edible portion, added ingredients, cut, and whether skin or bone is included.
The batch-yield method
For a mixed recipe, record the ingredient amounts in the same states used by their database entries. Add their nutrients to get the recipe total. After cooking, weigh the finished edible batch. Nutrients per gram equal total recipe nutrients divided by finished batch grams.
If a finished pot weighs 1,200 g and your bowl is 300 g, your bowl is one quarter of the batch. This method automatically accounts for the batch’s water gain or loss without inventing a correction factor.
What about leftovers?
Cooling does not change the batch total simply because the scale reading shifts slightly. Evaporation during storage can change weight, and sauces can settle unevenly. Stir mixed dishes before portioning when practical and treat home calculations as estimates, not laboratory measurements.
Consistency beats false precision
A five-gram difference in a database match can matter less than switching randomly between raw and cooked entries. Choose a repeatable method, save the exact entry, and use the same process when comparing weeks. For ordinary meal planning, a defensible estimate is more useful than extra decimal places built on mismatched data.