Every entity in the dataset, with how much of it there is.

The catalogue is unusable without this. A buyer can read the schema and still not know whether the node they care about is in here, how far back it goes, or what it last cleared at — and the answer to "what can I reference?" is the first question anyone asks of a feed.

Aggregated per node rather than sampled, so the counts are the real ones — once per collector run, not once per request (entities.py). Coverage and facets describe the whole table whatever the filter, because a tier count that shrank to match the search would be answering a different question.