Praudhamanorama p145

p145

As for what Haradatta said, that even under the view of analysis, this [rule] is meant for particles and onomatopoeic words—that is questionable, because even there the assumption of a base and so forth is possible.
In reality, however, even under the view of derivation, the rule is meant for cases such as bahupaṭavaḥ. Nor can it be objected that in that case one should simply formulate the rule as "bahucpūrvaḥ"; because in cases like mūlakenopadaṃśam there would be an overextension due to being an ending in a kṛt affix. Nor is there a way out by saying that the mention of samāsa is meant as a restriction, because without the formulation of the rule containing arthavat, the mention of samāsa cannot serve as a restriction. This is the direction.
[Under the rule] "kṛttaddhitasamāsāśca" (Pāṇini 1.2.46), "arthavat" is continued; on the strength of that, the rule applies to that which ends in them. For in a semantic integration where the constituent loses its own meaning, there is no meaning for kṛt and taddhita affixes. Nor is recourse to a former status commended. Intending that even in an integration where the constituent does not lose its own meaning, the meaning does not culminate in the affix alone, he says: "ending in kṛt and taddhita". When the exclusion [of prātipadika status] would otherwise apply by "adhātuḥ" in liṭ, dhuk, and so forth, and by "ending in a non-affix" in kartā and so forth, this rule is an enjoining rule.
To the phrase "for the sake of restriction": it should not be doubted that when an exclusion would apply due to ending in an affix, let it be an enjoining rule; for a sentence does not end in an affix, because when an affix is mentioned, only that which begins with what it was enjoined after is grasped. If it is objected: is not the restriction to that which begins with it prohibited in the case of a feminine affix, and thus in rājakumārī and the like, let it be an enjoining rule? [We say:] do not say so, because even an internally conditioned elision by halṅyābilopa does not apply in the domain of luk, and compound formation takes place while the case ending sup is still heard; and there, the technical name which has applied as ending in a non-affix is easily available by the maxim of that which is altered in one part even when the case ending has undergone luk. For this very reason, the augment num and others do not occur in gomatpriya and the like; whereas if an internally conditioned operation applied prior to luk, they would certainly occur.