since 'data' does not (accessibly) exist, but is only filtered through observations operationalized according to frames of interpretations (including measurements), will/can/should one always know
(a) that the data fails to fit the theory and(b) that the particular construction of the data is not what needs changing rather than the 'theory'?
 To be clear, here Bateman locates 'data' outside language; that is: as the experience that is construed as meaning by language. On the SFL model, the data to be modelled as linguistic theory are the construals of experience as language.
 To be clear, testing the fit between data and theory is testing the fit between
- construals of experience as language and
- construals of language as linguistic theory.
 To be clear, a "particular construction of the data" is a construal of language in terms of a linguistic theory. If the theory is inaccurately applied to the data, then it is the "particular construction of the data" that needs changing. If an accurate application of the theory to the data reveals a shortcoming in the theory, then it is the theory that needs changing.
For an analysis of a previous example of Bateman's epistemology, see John Bateman Denying The Existence Of Text.