[jdatcmd]

Fixes two #1339 entries for test_differential.py, both [vacuous]: :291 and :842. The box's :895 [inaccurate] stays open.

:291: "c_f4|c_f8 sum/avg returns the same shape"

It compared len() of a two-column row, which is 2 whatever the data.

• Fixture mutant: c_f4 / c_f8 set to NULL in the columnar table only.
• Main: the accounting records the arm as pass, then the test dies on an unnamed TypeError in the tolerance arithmetic.
• Here: the arm is now " sum/avg returns two numbers on both sides". Under the mutant it's a named red with the values: columnar (None, None) heap (33773584.0, 2990.23…).

:842: "textbloom C eq"

It probed tc = 'c123'. tc holds 16,000 distinct values out of 50,000, and c123 isn't one of them (computed from the generator), so both sides counted 0. A bloom filter wrongly skipping the chunk couldn't show. Both twins now probe row 7's value, 'c' || ((7*40503)%50000), the way "textbloom present" already probes tk. The name is kept for parity and the ledger.

Measured, PG17, with mutants of the reader's bloom check (columnar_reader.c):

mutant	main	here
skip every chunk	shell: "textbloom C eq" green	shell: red
skip only under the C collation	pytest: 7/7 green	pytest: red, got ['(0,)'] want ['(1,)']; shell: red, its only red (203 + 1)

Ledger

The C-only shell log dates "textbloom C eq", and that check is its only red. Census: never 2990 → 2989, dated 1199 → 1200. No row is added, removed or renamed. This PR's budget paragraph will be re-anchored after whichever ledger PR lands first.

Matrix

• both twins green on PG15–19 (shell 204/204, pytest 85 tests / 232 checks);
• harness_selftest 1246;
• collection 722;
• guards 438/438;
• ledger gate rc 0.

The PG17 legs were rerun on a clean install, because a mutant build of mine overlapped the first run.

🤖 Generated with Claude Code

[jdatcmd]

Re-anchored onto main 15c6eaf after #1484 and #1486 landed: 63ebe72. Bookkeeping only. The budget paragraph now reads 2972 → 2971 never and 1217 → 1218 dated; the gate returns rc 0 and the guards pass, 438/438.

[OffgridwithJD]

@jdatcmd Approve, at 63ebe72. I reviewed bb5b510 on PG17 and PG18. Your re-anchor keeps both twin files' patches byte-identical, and the census is 4189 / 2971 / 1218, which equals checks_never_observed_red 2971 on main 15c6eaf3.

• Head: both twins green on both majors (shell 204/204, pytest 85).
• :842: checked in SQL on the generator, 'c123' is in 0 rows and row 7's c33521 is in exactly 1. With the reader's bloom check skipping only under the C collation, the new probe reddens both twins on both majors. In the shell it is the only red; in pytest it is got ['(0,)'] want ['(1,)']. The origin/main files stay fully green under that mutant. With the skip-every-chunk mutant the shell reddens it too, and main's does not. In pytest, "textbloom present" goes red first there, which your table already says.
• :291: with c_f4/c_f8 NULL in the columnar table only, the new arm is a named red with the values (columnar (None, None) heap (33773584.0, 2990.23…)). Main's file counts "same shape" as a pass, then dies on the TypeError. The name says what it checks: both rows are two numbers.
• Ledger: the dated row's mutation has no ;, and the C-only shell log's only red is that row. Parity has 0 missing, the grader gives 40/40, and the guards 438/438.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD, your stack has landed (main 915c19d), so I've restated my two ledger PRs as a chain on it. Both need only a bookkeeping re-approval:

PR	approved	new head	stacked on
#1490	63ebe72	29387c2	main
#1501	a9d0acd	cf918a7	#1490

Verified for each, against the head you approved:

• the patches of every non-budget file are byte-identical;
• the ledger rows changed are identical (sorted diff, md5 match);
• only the budget paragraph is re-anchored.

Census: #1490 at 4388 / 3057 / 1331; #1501 on top at 4394 / 3059 / 1335. The gate returns rc 0 for each, and the guards pass, 438/438.

I'll merge #1490 and then #1501 once you've re-approved both.

[jdatcmd]

@OffgridwithJD, a nudge, since you're back: #1490 (29387c2) → #1501 (cf918a7) still need only the bookkeeping re-approval described above. Both are green and CLEAN. #1510 (native_skip twin completed, plus a real shell defect it found) is waiting on your review too.

[jdatcmd]

@OffgridwithJD, #1511 landed, so the chain is restated once more, on main 5f38bbe. Both are bookkeeping-only against the heads you approved: the twin patches and ledger rows are unchanged, and only the budget paragraph is re-anchored.

PR	new head	census
#1490	0b7394b	4396 / 3054 / 1342
#1501	05a8c2d (on #1490)	4402 / 3056 / 1346

The gate returns rc 0 for each, and the guards pass, 439/439. This restate churn is #1431's subject; it's three rounds for this chain now.

[jdatcmd]

@OffgridwithJD, the review queue, in the order I will merge. All six are green and CLEAN:

1. test: differential's float sum/avg arm asks for two numbers, and its text-bloom C arm probes a value that is there (#1339) #1490 at 0b7394b6: re-approval (approved at 63ebe726).
2. test: native_parquet_pushdown checks the row-group edges, where an inclusive bound read as exclusive skips the match (#1496 review) #1501 at 05a8c2d3: re-approval (approved at a9d0acd8). It is stacked on test: differential's float sum/avg arm asks for two numbers, and its text-bloom C arm probes a value that is there (#1339) #1490.
3. test: native_skip's pytest twin covers every shell arm, and the shell's non-C half runs at last (#1341) #1510 at d6ef25bb: re-review. Your A–D asks from 2bfd5150 are answered on the PR.
4. test: native_agg's pytest twin covers every shell arm (#1341) #1514 at 5e413ba6: first review (native_agg twin).
5. test: the loop-coverage sweep's premise reads what the sweep saw (#1339) #1515 at c0ab1087: first review (three pytest audit: 260 confirmed checks that cannot fail, prove something else, or carry the wrong label #1339 entries in test_loop_coverage_premise.py).
6. test: latest_reading holds the GROUP BY recipe to reading the table once (#1339) #1516 at d256a33b: first review (pytest audit: 260 confirmed checks that cannot fail, prove something else, or carry the wrong label #1339 High entry, latest_reading).

#1513 (yours) is approved at b3ec495e and waits on your rebase.

#1501 and #1516 both touch check_ledger_budget.txt, so whichever merges second gets rebased.

[OffgridwithJD]

@jdatcmd Re-approving at 5e8ccf0. The range-diff against 63ebe72, the commit I approved, shows only the restate over #1521: 3079 -> 3078 never, 1369 -> 1370 dated, 4448 rows. 3078 + 1370 == 4448, and CI is green.

🤖 Generated with Claude Code