dimension	value	meaning
join_type	inner	JOIN
join_type	left	LEFT JOIN (NULL-extended rows are counted)
join_type	right	RIGHT JOIN
join_type	full	FULL JOIN
join_type	semi	IN / EXISTS
join_type	anti	NOT EXISTS / NOT IN / LEFT JOIN .. IS NULL
cardinality	1:1	one row on each side per key
cardinality	N:1	many fact rows per dimension row (the common lookup)
cardinality	1:N	a parent with many children, queried from the parent side
cardinality	M:N	through a link table
size_relation	tiny_dim	a dimension of 3 rows
size_relation	small_dim	a dimension of about 1/1000 of the fact rows or less
size_relation	mid_dim	a dimension of about 1/10 to 1/100 of the fact rows
size_relation	dim_eq_fact	a dimension as large as the fact table
size_relation	fanout_1000	about 1000 fact rows per dimension row
key_type	int	int4 join key
key_type	bigint	int8 join key
key_type	text	text join key
key_type	uuid	uuid join key
key_type	date	date join key
key_type	composite	two or more columns
key_type	expression	an expression of columns (lower(email), a cast, a date_trunc)
query_kind	group_by_dim_attr	GROUP BY a column of the joined table
query_kind	filter_dim_attr	a condition on a column of the joined table
query_kind	count_distinct_across_join	count(DISTINCT x) over a join
query_kind	sum_avg	sum / avg of a fact measure through the join
query_kind	filter_agg_dim_attrs	aggregate FILTER (WHERE) over dimension attributes
query_kind	top_n_across_join	ORDER BY .. LIMIT over a join
query_kind	latest_per_group	the latest row of each group (DISTINCT ON, row_number, LATERAL)
query_kind	exists_scope	EXISTS over an assignment table that scopes the rows
query_kind	anti_join_count	rows with no match
query_kind	chain	a chain of 3 or more tables (child -> parent -> grandparent)
query_kind	star	a fact with 3 or more dimensions
query_kind	snowflake	dimensions that have dimensions
query_kind	self_join	a table joined to itself
query_kind	range_join_versioned	a range (valid_from, valid_to) or inequality join to a versioned table
query_kind	missing_field_count	counts of rows whose attribute is NULL or blank
query_kind	lookup_exact_key	point lookup by an exact key through the join
query_kind	bm25_then_join	a BM25 text filter, then the join
query_kind	jsonb_attr	a jsonb attribute joined to a catalog
query_kind	partition_prune	a partitioned fact pruned by a dimension condition
query_kind	set_algebra	any / all / none of a set (tags)
query_kind	graph_edges	edges between targets of one table (follows, mutual)
write_pattern	append_only	insert batches only
write_pattern	dim_attr_churn	updates that change a dimension attribute, so the derived class of many facts changes
write_pattern	key_churn	join keys change on the fact or the dimension side
write_pattern	delete_heavy	large deletes
write_pattern	delete_resurrect	deleted rows come back
write_pattern	late_arriving_dim	a dimension row arrives after the facts that name it
write_pattern	orphan_fact	facts whose key has no dimension row
write_pattern	bulk_reload	a large load or reload in one statement
write_pattern	soft_delete	a flag or timestamp marks rows deleted, rows stay
write_pattern	maintenance	VACUUM / REINDEX CONCURRENTLY in the middle of the workload
scale	S	100,000 fact rows
scale	M	1,000,000 fact rows
scale	L	10,000,000 fact rows
skew	uniform	keys drawn uniformly
skew	zipf	power-law key frequencies (skew knob > 1)
skew	hot_key	one key holds a fixed share of the facts (hot knob)
clustering	random	join key independent of heap order
clustering	clustered	heap order follows the join key
clustering	anti	heap order falls as the join key grows
storage	heap	heap tables (the default)
storage	sift_table	the fact is a sift_table (column store access method); the native engine uses a heap table
query_kind	parent_exists_child	EXISTS / IN over the children of a parent with a condition on the child, seen from the parent
query_kind	parent_not_exists_child	NOT EXISTS / LEFT JOIN .. IS NULL: parents with no child that matches a condition
query_kind	parent_left_agg	parent LEFT JOIN child with count / sum / count(DISTINCT) grouped by a parent column
query_kind	child_count_per_parent	number of children of each parent, grouped by a parent column
query_kind	top_n_per_parent	the best N children of each parent (LATERAL .. LIMIT, row_number), aggregated by a parent column
query_kind	parent_child_count	a condition on the number of children of a parent (count(*) >= n)
query_kind	child_agg_per_parent	an aggregate of the children of each parent (max, sum, count) joined back to the parent
query_kind	semi_join_count	rows with a match counted (EXISTS, IN, a join that only counts), no column of the matched table read