diff --git a/sql/index_candidates_to_partial.sql b/sql/index_candidates_to_partial.sql
new file mode 100644
index 0000000..387ed6c
--- /dev/null
+++ b/sql/index_candidates_to_partial.sql
@@ -0,0 +1,19 @@
+--list single-column indexes which have only 5 or less distinct values in 95+% of rows with not null values.
+--so they possibly may be replaced with partial indexes to save disk space.
+
+SELECT
+pg_index.indrelid::regclass AS table,
+pg_attribute.attname AS column,
+pg_index.indexrelid::regclass AS index,
+pg_stats.null_frac,
+pg_stats.most_common_freqs[1:5] AS most_common_freqs,
+(pg_stats.most_common_vals::text::text[])[1:5] AS most_common_vals,
+pg_size_pretty(pg_relation_size(pg_index.indexrelid)) AS index_size,
+pg_get_indexdef(pg_index.indexrelid) AS index_def
+FROM pg_index
+JOIN pg_attribute ON pg_attribute.attrelid = pg_index.indrelid AND pg_attribute.attnum = ANY(pg_index.indkey)
+JOIN pg_stats ON (pg_stats.schemaname || '.' || pg_stats.tablename)::regclass = pg_attribute.attrelid AND pg_stats.attname = pg_attribute.attname
+WHERE pg_relation_size(pg_index.indexrelid) > 10*8192
+AND (SELECt SUM(a) FROM unnest(most_common_freqs[1:5]) a) >= 0.95
+AND array_length(pg_index.indkey, 1) = 1
+ORDER BY pg_relation_size(pg_index.indexrelid) DESC,1,2,3;