The laws of definition are:
Honesty: do not contradict within definitions
Common ground: definitions contain other definitions; they involve comparing patterns
Agreement: definitions involve agreement
Honest agreements creating common ground: definition.
"The Kingdom Of Heaven is like a tree, the birds of the air come and rest in its branches".
See recent topics on quantum electrodynamics unplugged, and on Dr. Dick's equations 1.1 and 1.2.
"I can define anything to be anything else all I want."
This involves MATCHING the first "anything" with the "anything else".
I call these two "anythings": musical chairs games. Dr. Dick calls them "adding unknown data".
"But if I then derive a theory..."
I call this "comparison of the two patterns", or a "third musical chairs game that joins the dots of the other two".
".. and predictions of the theory do not agree with experiment"
"experiment" involves definitions that define it. So this is a fourth musical chairs game.
"predictions" are guesses. Where the "theory" is logically consistent with the definitions that supply it; it is a valid way of looking at the known data perhaps. In so far as it is built on unknown data; it is guesswork (or it is a broad probability true by definition within the margin of error).
"predictions" are either "true by logical consistency with your definitions"; or involve unknown components and guesses. Possibly "prediction" is an illusion.
", then I must assume that the definitions are unrealistic"
True; the failure of the prediction might tell you that the "definitions" were logically inconsistent, or contained gaps or unknown data.
If you are precise with definitions, you can try to identify the error margins that govern them. Then you can make accurate so-called "predictions"
that are probabilities true by definition. Example: quantum electrodynamics.