[OffgridwithJD]

@jdatcmd The sixth #1339 custom-scan audit entry, done as a twin PR because sorted_projection.sh had none. Stacked on #1504 (#1503 → #1504 → this). This PR's own commits are the top two.

Three shell defects, each measured red first

arm	mutation	old suite	new suite = twin
"index point lookup after sort"	post-sort reindex removed (columnar_vacuum.c), so the index points at the old row numbers	green: it never read the index	red
"reject non-columnar table"	vacuum_sorted's columnar check removed	green: the heap table still raised, as XX001 reading the heap's file as columnar storage	red
any expect_error arm going red	(by construction)	recorded as "<label> (expected error)", a name the passing run never records, so its ledger row could never be dated	recorded under the label

Fixes:

• The index lookups turn the custom scan off, with a premise each that they read t_col_id_idx. The count plans as an Index Only Scan, so the premise accepts either index node.
• Each refusal requires its own SQLSTATE: 42809, 42703, 22023 and 42883, measured at head.

The twin

All 29 checks by the shell's name, in three tests, porting pgc_check_ordered_oracle and the set and sequence hashes. Under each mutation (reindex removed, no sort, columnar check removed) it reddens exactly the shell's set. The no-sort mutation also reddens the json refusal, because that refusal is raised by setting up the sort.

Five-major runs at head: 29/29 for the shell and 29/29 for the twin, on each of 15–19.

Ledger

Census 4377 / 3053 never / 1324 dated (on #1504's 4348 / 3029 / 1319), with suites_not_covered 192 → 191. 5 of the 29 checks are dated. The gate gives rc 0. Collection gives cluster_tests 744 -> 747, and TESTS.md gets §174. The loop-reader list gains the suite, the grader passes 40/40, and the guards 438/438.

That leaves audit.sh as the last audit entry.

🤖 Generated with Claude Code

[jdatcmd]

Approve, at 332435e (its own commits, on #1504).

My own mutant, different from your three: vacuum_sorted's unknown-column refusals raise 22023 instead of 42703 (all four ERRCODE_UNDEFINED_COLUMN sites in columnar_vacuum.c). PG17:

• shell: reject unknown column: got [22023] want [42703];
• pytest: the same.

That's what the per-refusal SQLSTATE buys. The old "any error" form would have stayed green under it. The (expected error) naming fix, recording a red under its own label, is the right one too: a name the passing run never records can't be dated.

[jdatcmd]

@OffgridwithJD, merging #1506 now: approved at 332435e, 15/15 green, CLEAN.