[OffgridwithJD]

#1341, the twin-less suites: native_decode_gating.sh (#452 phase 2) had no pytest twin. This PR adds one, every check by the shell's name, and seeds the suite into the ledger. The port found that two of the filter counters' guarantees couldn't fail, and that the multi-group fixture depended on the gating it was built to test. Both are fixed in both twins.

Stacked on #1455, at the top of my chain. This branch is #1455's head plus two commits.

What changes

• test/pytest/test_native_decode_gating.py, the twin: two tests in a private schema. The width gate (Decode gating (#452 phase 2) is unconditional; gate it on selectivity so it stops regressing less-selective scans #595) is set off for the session, where the shell pins it off for its cluster; it's PGC_USERSET. compare_to_bash.py grades it as covering all 29 names, and it joins COMPLETE.
• Rejections inside a surviving vector weren't tested. Every row of the needle's surviving vector matched, so the producer's re-test of that vector rejected nothing. A producer that counted those rejections a second time therefore read exactly right.
  • Measured: with the non-counting filter disabled, the suite stayed 26/26.
  • The fix: a qual that half of vector 0 passes (tag LIKE '%needle%' AND p1 % 2 = 0, still no skip predicate) asserts its rejections are counted once, with a premise for its 512 survivors in one vector.
• The counters were checked only on the table without deletes.
  • Measured: with the evaluator's deleted-row test removed, the suite stayed 26/26.
  • The fix: Rows Removed by Filter is now checked on the multi-group table with deletes, against heap's live count.
• The fixture depended on the gating under test. Its DELETE … WHERE p1 % 17 = 0 is an unprunable qual too, so a gating mutant broke the fixture, and with it the premise "rows really were deleted", rather than the arms. pgc_ledger.py gate then refused, rightly, a premise row with a red observation.
  • The fix: the DELETE now runs with the width gate out of reach (100000, the setting's maximum), and the premise counts the survivors with count(*).
• Ledger: native_decode_gating is seeded with 29 rows from five-major runs, and 11 of the 29 rows dated. suites_not_covered goes 215 → 214, and the census is 3691 + 29 = 3720 rows, 2705 + 18 = 2723 never and 986 + 11 = 997 dated, matching the tsv. The gate gives rc=0 against origin/main, and no premise is dated.

Red first

Each mutation of src/columnar_reader.c was run one at a time on PG18, through both twins, with the .so md5 printed every time.

mutation	native_decode_gating.sh	test_native_decode_gating.py	the arms that went red
none (the head)	29/29	29/29	none
phase 2's per-vector evaluation never runs	25/29	1 of 2 tests fails	the decode counts (got [192] want [32]/[37]) and the skipped-vector counts
the evaluator counts a deleted row as filtered	28/29 (26/26 before this PR)	1 fails	rows removed by filter across deleted rows count only live rows: got [49140] want [46250]
every vector marked skipped	22/29	2 fail	the control's decode count, the needle's counts, and every correctness arm (got [EMPTY])
the producer counts a gated group's rejections a second time	27/29 (26/26 before this PR)	2 fail	rejections inside a surviving vector are counted once: got [32768] want [32256], and the deleted-rows counter

Never red (18):

• the thirteen premises;
• the observable's presence;
• the full scan across every group;
• the nomatch query's empty answer;
• two exact counters that no single mutation here moves.

Runs

major	native_decode_gating.sh	test_native_decode_gating.py
15	29/29	29/29
16	29/29	29/29
17	29/29	29/29
18	29/29	29/29
19	29/29	29/29

Also measured: the guard half (438 tests), harness_selftest 1245/1245 and docs_style. cluster_tests goes 674 → 676 by collection, and TESTS.md gets §148.

🤖 Generated with Claude Code

[jdatcmd]

Approving 51b5aa3.

Measured on PG17: native_decode_gating.sh 29/29; test_native_decode_gating.py 29 checks over 2 tests.

The deleted-row mutant, run myself: if (deleted) continue; removed from the phase-2 evaluator in columnar_reader.c, so a deleted row is counted and asked like a live one. Both twins go red on the new arm, and only on it: rows removed by filter across deleted rows count only live rows, got [49140] want [46250]. Your table says the same, and before this PR the counters were checked only on the table without deletes.

I took the other three mutants from your table. This goes on top of the stack I approved (#1450 → #1455); the stack's restate onto 0edb5ae covers it too.

[jdatcmd]

Re-approving fb1f6e2, the restate onto main db56168. I compared every commit's patch with the approved stack using git range-diff. The test and source commits are unchanged, and the ledger-seed commits changed only by re-anchoring on #1470. The census at this head matches its budget paragraph, and the counts step consistently. At #1467's head: the guards pass (438), 687 collected (matching the file), and harness_selftest is 1245/1245.