From a science and engineering point of view:

If you want it your theory to be compatible with "reality" , you use all data available, to the limit of available precision. If there are data points it doesn't fit, you need to be able to characterize those and explain whether they are erroneous data, systematic exceptions which have some pattern to them, or something else.

If the theory only works with a subset of that multivariate space, it may still be useful for that subset but is not a complete model of reality. Science advances by understanding the limits of its theories and refining them or replacing them as ways are found to make the theory match in more situations and to greater precision. Science never presumes to have the ultimate answer to complete precision, it is just an ongoing set of best available models.

If you are talking about some other kind of truth and reality, I am the wrong person to try to answer.

From a science and engineering point of view:

If you want your theory to be compatible with "reality" , you use all data available, to the limit of available precision. If there are data points it doesn't fit, you need to be able to characterize those and explain whether they are erroneous data, systematic exceptions which have some pattern to them, or something else.

If the theory only works with a subset of that multivariate space, it may still be useful for that subset but is not a complete model of reality. Science advances by understanding the limits of its theories and refining them or replacing them as ways are found to make the theory match in more situations and to greater precision. Science never presumes to have the ultimate answer to complete precision, it is just an ongoing set of best available models.

If you are talking about some other kind of truth and reality, I am the wrong person to try to answer.