# shape	lane	rows	chosen	cost	best	cost	M instr chosen	M instr best	ratio	state	note: the chosen plans that lose to the best by over 20% and 1M instructions (scripts/plangate.sh --calibrate --update)
search_one_parent	stock	100000	bitmap=native	766.1	index	3086.4	4.71	1.81	2.60	known	BitmapAnd of events_memo (62,500 TIDs of 250,000 in 1.45 to 1.69 ms, 25.6 ns a posting, priced 25.77) and events_col (486 rows); the sift_col index scan with the memo as a filter reads the same 410 heap pages at random_page_cost (est 3,211 against the bitmap's 875). Pricing the posting at its measured 5.8 operator costs (POSTING_CPU_FRACTION 1/16 to 5.8) prices the memo node at 922.25 and the bitmap plan at 1,766.7 and the pick does not move, while search_count (tuned, 250,000 rows) became wrong (scan 52.55 M instr against 0.03 M): reverted. The cost left is PostgreSQL's page price of cached heap pages
search_one_parent	stock	250000	bitmap=native	901.7	index	3242.9	12.90	2.18	5.92	known	BitmapAnd of events_memo (62,500 TIDs of 250,000 in 1.45 to 1.69 ms, 25.6 ns a posting, priced 25.77) and events_col (486 rows); the sift_col index scan with the memo as a filter reads the same 410 heap pages at random_page_cost (est 3,211 against the bitmap's 875). Pricing the posting at its measured 5.8 operator costs (POSTING_CPU_FRACTION 1/16 to 5.8) prices the memo node at 922.25 and the bitmap plan at 1,766.7 and the pick does not move, while search_count (tuned, 250,000 rows) became wrong (scan 52.55 M instr against 0.03 M): reverted. The cost left is PostgreSQL's page price of cached heap pages
search_one_parent	tuned	30000	index=native	3025.1	bitmap	3381.8	4.98	1.35	3.69	known	table lane tuned, 30,000 rows: the index scan 4.98 M instr (est 3,025), the bitmap 1.35 M instr (est 3,382); the BM25 posting price was tried at the 250,000-row rows of this shape and not kept
search_one_parent	tuned	250000	bitmap=native	3232.5	index	3530.3	10.96	2.18	5.03	known	BitmapAnd of events_memo (62,500 TIDs of 250,000 in 1.45 to 1.69 ms, 25.6 ns a posting, priced 25.77) and events_col (486 rows); the sift_col index scan with the memo as a filter reads the same 410 heap pages at random_page_cost (est 3,211 against the bitmap's 875). Pricing the posting at its measured 5.8 operator costs (POSTING_CPU_FRACTION 1/16 to 5.8) prices the memo node at 922.25 and the bitmap plan at 1,766.7 and the pick does not move, while search_count (tuned, 250,000 rows) became wrong (scan 52.55 M instr against 0.03 M): reverted. The cost left is PostgreSQL's page price of cached heap pages
spine_dims_cross	tuned	30000	scan	12491.1	seq	265447.5	183.29	114.93	1.59	known	table lane tuned, 30,000 rows: the chosen scan 183.29 M instr (est 12,491), the sequential scan 114.93 M instr (est 265,448): the profile's seq_page_cost 100
spine_dims_cross	tuned	250000	index=native=scan	55786.1	seq	1145526.0	1008.19	567.34	1.78	known	tuned, 250,000 rows: the chosen plan 1,008.19 M instr (est 55,786), the sequential scan 567.34 M instr (est 1,145,526): the profile's seq_page_cost 100 prices the sequential pages that are cached
spine_expression_key	tuned	30000	index=native	12272.5	seq	262847.6	269.61	95.85	2.81	known	table lane with seq_page_cost 100, 30,000 rows: the chosen index scan 19.7 ms, 144 blocks, 269.61 M instr (est 12,272); the sequential scan 10.0 ms, 31 blocks, 95.85 M instr (est 262,848): the profile prices a sequential page at 100 and the pages are cached; decoupling the 2x decode pages from seq_page_cost moved no pick and 83 band rows (reverted)
spine_range_join	tuned	30000	index=native	14019.1	seq	267877.2	369.17	178.53	2.07	known	table lane with seq_page_cost 100, 30,000 rows: the chosen index scan 24.8 ms, 2,015 blocks, 369.17 M instr (est 14,019); the sequential scan 15.5 ms, 31 blocks, 178.53 M instr (est 267,878): the same page price of the profile as spine_expression_key
top_rows_per_kind_unindexed_column	tuned	30000	scan	5857.3	seq	267251.5	495.86	396.08	1.25	known	table lane tuned, 30,000 rows: the chosen scan 495.86 M instr (est 5,857), the sequential scan 396.08 M instr (est 267,252): the profile's seq_page_cost 100
top_rows_per_kind_unindexed_column	tuned	250000	index=native=scan	104527.0	seq	1190935.0	4221.48	3306.54	1.28	known	tuned, 250,000 rows: the chosen plan 4,221.48 M instr (est 104,527), the sequential scan 3,306.54 M instr (est 1,190,935): the same seq_page_cost 100 of the profile