I have always perceived an inherent contradiction between Chomsky's 'no tampering' idea and ANY version of Merge (or any Merge-like operation) driven - under the principle of Economy - by the need for features of a syntactic object A to be 'checked', 'satisfied', etc. by 'properties' (features, values) of a second object B. Whether the checking/satisfaction process involves unvalued features becoming valued, or valued features becoming 'licensed', or 'agreeing' between themselves, etc. is inmaterial, as far as I can see, at this level of generality.
Not that any sophisticated reasoning is involved on my part in this respect, on the contrary: I just feel that if, say, an unvalued attribute of a category A gets valued as a consequence of A's being Merged to B (or a valued attribute of A is 'licensed' by, say, 'agreement' with a parallel attribute of B, etc.), then A, B, or both, have been 'tampered with' in a flagrant way. After all, a 'category' (a syntactic object visible to Merge) being ultimately just a set of such 'features', to the extent that the state of whatever attribute-value pair of A or B may be involved is altered by the merger, so is the object A (or B) itself.
I do not know whether 'no tampering' remains an important principle of current minimalist syntax, but, if it does, I would like to know how that difficulty (if it is such and not just a misconception of mine) has been circumvented, and, if it no longer does, I would still want to know why the idea has been abandoned.