Why track catalogue coverage, diversity and novelty when the relevance metrics are already rising?

Because relevance metrics also rise for a degenerate system. A model that shows everyone the top-100 bestsellers scores decent Precision and NDCG — popular items really do get clicked. Meanwhile the catalogue stops selling, new items never get a chance, and every feed looks identical. Coverage captures what share of the catalogue is ever surfaced; diversity how dissimilar items are within one list; novelty how non-obvious the recommendations are for that user. These are guardrail metrics: keep them inside a corridor while optimising relevance.