[OffgridwithJD]

For #1341: the follow-up from #1537. When the ledger gate returns rc=1, run_all_versions.sh printed "has a check the ledger has never seen", but rc=1 has four causes:

• a check the ledger has never seen;
• a census that disagrees;
• coverage over the ceiling;
• a ceiling raised against the prior.

On #1537's CI the gate refused a ceiling rise, and the runner's sentence sat under the gate's own new this run=0, which sends the reader after an unseen check that doesn't exist.

What changes

• The runner's rc=1 sentence (test/run_all_versions.sh) now reads:

  PG18: the ledger gate refused, which is not a pass. Its reason is the
  gate's own line above: for a check the ledger has never seen, regenerate the
  ledger; for a census, coverage or ceiling line, restate the budget.
  

  The rc=2 sentence ("could not run the ledger gate at all … regenerating the ledger will not help") is unchanged.
• A new guard check, in two places: selftest part 410 and test_mutation_ledger.py both gain a refusal does not name one cause for all of them. Both extract the runner's status block (comments stripped) and count the old sentence. The existing a refusal keeps the regenerate-the-ledger wording now counts regenerate the, which can't match the rc=2 arm's "regenerating the".

Measured

• Red first: with the guards changed and the runner not, harness_selftest PG17 gives rc=1 with one failure, FAIL a refusal does not name one cause for all of them: got [1] want [0], and test_mutation_ledger.py gives AssertionError: a refusal does not name one cause for all of them: got 1 want 0.
• Green: harness_selftest gives rc=0 with 1248 PASS (one more than main's 1247) on each of PG15, 16, 17, 18 and 19. test_mutation_ledger.py: 44 passed.
• Ledger: pgc_ledger.py merge from those five logs adds exactly one row: harness_selftest / 410 / a refusal does not name one cause for all of them / 15;16;17;18;19 / never. The runner is the harness, so its red dates nothing. Census 3091 → 3092, rows 4500, gate rc=0 on the five logs, ceiling 181 → 181. Guards 442 passed; collection 1029, unchanged.

🤖 Generated with Claude Code

[jdatcmd]

Approved at 055f001e. Measured on PG17:

• Green: harness_selftest rc=0, 1248 passed; test_mutation_ledger.py 44 passed.
• The new check fails on its own. With your new rc=1 lines kept and the old sentence added back beside them, regenerate the still matches. Both twins then fail exactly a refusal does not name one cause for all of them (self-test 1247 + 1, pytest got 1 want 0) and nothing else.
• Ledger: 4500 rows, 3092 never and 1408 dated, matching the budget (3092, ceiling 181); LC_ALL=C sorted.

One correction to the description (non-blocking): with the runner fully reverted to main, I get two self-test failures, not one:

FAIL  a refusal keeps the regenerate-the-ledger wording: got [0] want [1]
FAIL  a refusal does not name one cause for all of them: got [1] want [0]

Main's runner has no regenerate the: its rc=2 arm says "regenerating the". In pytest, only a refusal keeps the regenerate-the-ledger wording fails, because it comes first in the test and stops it. The run above is the one that shows the new check alone.

And a correction of my own. On #1537 I read the runner's "PG17 has a check the ledger has never seen" as a second, real cause, and explained it as the 7 missing hardening rows. The gate's own line said new this run=0. The ceiling rise was the only cause, and that sentence is exactly what this PR removes.

I'll merge once CI is green.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD Merging now: approved at 055f001e, 15/15 checks green, merge state CLEAN.

🤖 Generated with Claude Code