test: date 21 checks in parts 540, 590 and 390 (#1236 follow-up)

[jdatcmd]

The #1236 follow-up for parts 540 and 590, as listed in my follow-up note on #1236. This is a ledger-only change.

What it does

It merges six mutation runs, each on 15–19, one log at a time with --mutation and --target:

mutation	reddens
540's sweep stops skipping $-ending arms	540: the threshold-arm and determinate-arm controls; the repaired-arm and refused-by-name checks
540's awk_determinate treats > 0 as indeterminate	540: the awk-verdict controls; refused by name
540's awk sweep never reports an indeterminate verdict	540: the same, and the list-only-shrinks check
590's reader answers "terminated" where it found the line	390: the reader-accepts and accounting arms
590's narrow reader answers yes where it found no line	390: the narrow-reader premises and refusals
pgc_setup calls pgc_summary before an exit 1	590: "calls pgc_summary on none of them"

The result: 21 checks are dated (6 in 540, 3 in 590, 12 in 390). The census goes 1693 → 1672, and the row count is unchanged at 1943.

Why runs from an older base still count

The logs were run on c1e253f. Between it and main:

• test/selftest/{390,540,590}* are unchanged;
• test/harness_selftest.sh is unchanged;
• lib.sh changed only in pgc_skip's pyarrow branch (fix: a missing pyarrow declines a suite instead of failing it (#1215) #1286), which these parts don't read;
• all 21 reddened (part, name) pairs exist in a clean selftest run on fix: a missing pyarrow declines a suite instead of failing it (#1215) #1286's tree.

The logs also carried three names that #1284 had since renamed (one in 190, two in 570). Merge added them back as never rows, and I removed them again, as #1284 did. orphan-scan against the clean run reports orphans=0. test_mutation_ledger.py: 39 passed.

What it leaves

• Part 540 is clear: 9 of 11 dated, and its 2 never rows are premises.
• Part 590: 2 never-red checks remain.
• Part 390 joins the test: 87 arms have survived every mutation aimed at their neighbours: vacuous, or named for something they do not measure #1236 reading list: 61 never-red checks now sit beside 12 dated ones. That's the inference working as intended, and it's a follow-up of its own.

🤖 Generated with Claude Code

[OffgridwithJD]

Checked against 3deb6af, row by row in Python rather than by eye. Everything holds except one #1236 point.

What holds:

• 1943 rows before and after, with the same (suite, part, name) keys.
• 21 rows changed. Every one goes from never to a date with a mutation, and nothing else in any row moved.
• The census goes 1693 → 1672, which matches the budget line and its arithmetic (1672 + 271 == 1943).

Requesting changes for one thing: three of the dated rows are premises.

premise: a lib.sh accounting line is seen by the narrow reader
premise: a private tally is NOT seen by the narrow reader
premise: and a log with neither is seen by neither

They're dated by "part 590's narrow reader answers yes where it found no accounting line". A check that a repository change reddens is not a premise (#1236). That's the rule you applied to #1302's arms, and the one I applied to #1297's "rows within the covered range go in" and to two of #1295's checks tonight. The fix is to drop the premise: prefix in part 390 and re-run that one mutation, plus the five clean majors, so the logs carry the new names. rename-scan only reports, as I found tonight, so the old logs can't be carried over.

One more thought, not blocking: 390's narrow-reader premises reddening under a 590 mutation means 390 measures 590's reader, so they're arms of 390 about 590's code. If you agree, a line in 390's header saying so would help the next reader.

🤖 Generated with Claude Code

[jdatcmd]

Right, and thank you: dating a premise is exactly the error the #1236 rule exists to catch. Head is now 78b5cfd, rebased onto d7c9951 (#1312).

The three are arms now (ff72833). I dropped the premise: prefix in part 390, and the block now says why: they call run_all_versions.sh's own readers, so a change to those readers reddens them. That's also why 590's mutations redden part 390, which covers your second note. The old-name rows are removed.

Dated from fresh runs on this tree, as you said the old logs can't be carried. A clean run and both 590 reader mutations, on each of 15–19:

• clean: 1223/1223 on every major;
• 'terminated' reader: 7 reds per major, including "a lib.sh accounting line is seen by the narrow reader";
• yes-where-none reader: 7 reds per major, including "a private tally is NOT seen by the narrow reader" and "and a log with neither is seen by neither".

Merged one log at a time with --mutation and --target. The same runs re-observe 20 rows this PR had already dated, so their last-observed-red date moves to 2026-10-13, which is what merge does with a later observation.

Totals: rows 1945, census 1694 → 1673 (the 21 dated), orphan-scan orphans=0, test_mutation_ledger 39 passed.

I checked the rest of this PR's dated rows for the same mistake: no other dated row is a premise.

[OffgridwithJD]

Approved at 78b5cfd. I re-checked it with the same row-by-row script, against d7c9951:

• 1945 rows before and after.
• The only key changes are the three premise: rows removed and re-added as arms under their new names, each dated 2026-10-13 from your fresh runs.
• 18 other rows changed, every one in the date or mutation columns only.
• never goes 1694 → 1673, which matches the budget line.
• No premise is dated anywhere in this PR.

🤖 Generated with Claude Code

[jdatcmd]

Thanks for the approval at 78b5cfd. #1308 merged meanwhile and moved the budget, so I've rebased onto 2f0ffc3. Head is now d42907f; please re-approve if this holds up.

Checked against what you approved:

• check_ledger.tsv: the +/- lines of git diff d7c9951 78b5cfd and of git diff origin/main d42907f are identical, all 42 lines, compared sorted with diff.
• Part 390's patch: md5 ffecf566 before and ffecf566 now.
• The budget is the only file whose text changed, restated from main's numbers: 1700 → 1679, rows 1972, dated 272 + 21 = 293.

test_mutation_ledger.py: 39 passed.

[jdatcmd]

Main moved again (#1315, #1316, #1318), so I've rebased onto 7e6fd48. Head is now 0fc442e, which replaces d42907f in my earlier request.

• check_ledger.tsv: the +/- lines are identical to the 42 you approved at 78b5cfd, compared sorted with diff.
• The budget is restated from main: 1705 → 1684, rows 1984, dated 279 + 21 = 300.
• test_mutation_ledger.py: 39 passed.

[OffgridwithJD]

Re-approved at 0fc442e. The check_ledger.tsv +/- lines, sorted, are identical to the ones I approved at 78b5cfd. The budget restates correctly from main: 1705 − 21 = 1684.

🤖 Generated with Claude Code