[OffgridwithJD]

@jdatcmd Stacked on #1518, which is on #1517. Review only the top commit, e53ba5d.

The twin

test/pytest/test_native_bloom.py covers all 7 checks by the shell's name, one check per test.

• _skipped returns -1 when the plan has no "Chunk Groups Removed" counter, so "bloom off skips nothing" can't pass on a plan without the columnar scan.
• "bloom on skips groups" is at_least(..., 1) on the count.

The shell is unchanged: there an absent counter is an empty string, which neither skip arm accepts.

Evidence: the shell against the twin as shipped (raising expect), PG18

mutant	shell	twin	what goes red
none	7/7	7 passed	—
bloom consulted whatever the GUC says	1	1, the same	"bloom off skips nothing (min/max cannot)"
the GUC inverted (yours)	2	2, the same	"bloom on skips groups", "bloom off skips nothing"
bloom never consulted	1	1, the same	"bloom on skips groups"
bloom answers absent for every value	1	1, the same	"equality probe parity"
writer builds no bloom	2	2, the same	"bloom rows written", "bloom on skips groups"

The head run is 7/7 on each of PG15–19.

Bookkeeping

• Ledger: native_bloom seeded with 7 rows: 4433 → 4440, never 3071 → 3074, dated 1362 → 1366, suites_not_covered 187 → 186.
  • The 3 never are the row count, the custom-scan premise, and "absent value parity". Its value is outside every group's range, so min/max prunes it before the bloom is consulted.
• Gate: rc 0.
• Collection: cluster_tests 772 → 779, seven tests.
• Docs: TESTS.md gets §179, and the stem leaves unported_suites.txt.
• Guards: 440/440.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD, requesting changes at a7e42a8d for the same thing as #1518. In CI's harness, the first red in a test stops it, so the twin reports less than the shell.

My mutant: the bloom setting inverted

In columnar_reader.c:2053, pgcolumnar_enable_bloom_filter becomes !pgcolumnar_enable_bloom_filter, so the bloom is consulted when off and ignored when on. On PG17:

	red
shell	2: "bloom on skips groups" (got no), "bloom off skips nothing" (got 9)
twin	1: "bloom on skips groups" (got 0). 5 pass, 1 fail; the off arm never ran

Both arms sit in test_the_bloom_and_not_min_max_is_what_prunes, and the first one raises.

About "the soft twin" in your evidence tables. The tree has no soft mode: grep -i soft over pgc_vacuity.py and conftest.py finds nothing. So I read those tables as measured with expect patched locally not to raise. That shows the names can go red. It doesn't show what CI's twin reports, and that's the claim "redden the same names" makes. Could you take the red-set tables from the twin as shipped? Then the next defect of this kind shows up in your own table.

Suggested fix: one check per test, by parametrizing the cases as in #1518. Or, if you'd rather fix it once for every twin, a harness change that defers expect failures to the end of the test. That's a bigger question, and the owner should probably weigh in before the harness changes.

Everything else checks out: the -1 sentinel for a missing counter, the at_least on the skip count, and the never-reddened "absent value parity" explained by min/max.

[OffgridwithJD]

@jdatcmd agreed on all of it, and thank you for both mutants. I took (a), one check per test, in all three PRs:

PR	head	tests	how
#1517	03c31de	9	the module fixture runs every step in the shell's order and records each reading; each test asserts one
#1518	9bc605c	9	the reads and answers are parametrized by the shell's names (name,sql); the loop-reader entry is gone
#1519	e53ba5d	7	one function per check

The red-set tables now come from the twin as shipped, not from my local soft runner. You were right that the soft runner showed only that a name could go red. My runner now runs the real twin file and reads the AssertionError: <name>: got lines. Every mutant matches the shell, yours included:

• the by-reference null: 4 and 4 on PG18 and PG17;
• the inverted bloom setting: 2 and 2.

Also done:

• test: native_roundtrip gains its pytest twin, and asserts the read crosses its row groups (#1341) #1518: the row-group check drops premise:. It reads "the native table holds five row groups", and its ledger row is reseeded from new five-major logs.
• test: native_roundtrip gains its pytest twin, and asserts the read crosses its row groups (#1341) #1518: the TESTS.md text says the row count comes from metadata.
• The stack is rebased onto bbda341. The counts are unchanged, apart from cluster_tests (754 → 763 → 772 → 779).
• At each tip: gate rc 0, guards 440, and the collection count matches.

On (b), a deferred-failure mode for expect: I agree it's the owner's call. One-check-per-test fixes these three without touching the vacuity layer. I'll open an issue for it if the pattern keeps recurring in new twins.

🤖 Generated with Claude Code

[jdatcmd]

Approving at e53ba5d6. I took (a) in this PR and reran my mutant on the twin as shipped, comparing exact red-name sets from E AssertionError: <name>: got lines against the shell's FAIL lines. PG17:

mutant	shell red	twin red	sets
the bloom setting inverted (`columnar_reader.c:2053`)	2	2	identical: bloom on skips groups, bloom off skips nothing

Counts. One cluster_tests line, 779, and collection at this head is 779.

At the stack's tip (e53ba5d6), test_harness_deps, test_compare_to_bash and test_mutation_ledger pass 105.

I'll merge in stack order as CI goes green: #1517, then #1518, then #1519.

[jdatcmd]

@OffgridwithJD merging #1519 now at e53ba5d6: approved on this head, 15/15 checks, CLEAN.