[jdatcmd]

Fixes the #1339 entry test_tier_partition.py:213 [mislabelled], and one more check with the same problem.

What was wrong

"premise: the call fails with the fault injected after " reads the injected faults in tier_partition's own SQL (pgcolumnar--1.0-alpha5.sql). That is product code, so a repository change can redden the check: it's an arm (#1236).

Measured, PG17: I removed the five RAISE EXCEPTION 'injected fault after …' lines. All five step checks go red in the shell (got [ok] want [injected fault after <step>]), as does "premise: the first attempt failed after export and left its file", which reads the same fault.

The fix

• Both twins: the six checks lose the premise: prefix. There is no other change.
• Ledger: the six rows are renamed with their history. The same mutation reddens 17 checks in all, and all 17 are named as --target, because the merge refuses an unnamed red:
  • 16 rows that had never been seen red are now dated;
  • the already-dated retry arm gets the mutation appended to its history (;-separated, and its earlier entries are kept).
• Census: 2942 → 2926 never, 1160 → 1176 dated. No row is added or removed, and no key or major changes.

Measured

• Mutant, PG17: shell 17 red; pytest red on the first two (the rest masked, because expect raises on the first red).
• Matrix:
  • both twins 80/80 on PG15–19;
  • harness_selftest 1245/1245;
  • collection 710;
  • guards 438/438;
  • ledger gate rc 0, census agrees.
• The one skip: PG15–17 skip one check that PG18–19 run (79 + 1 skipped). Main's suite does the same; this PR doesn't change it.

🤖 Generated with Claude Code

[OffgridwithJD]

@jdatcmd Approve, at edd94b8.

• Head is green: both twins pass on PG17 and PG18 (shell 80/80, pytest 4/4).
• Your mutant matches the ledger: with the five injected-fault RAISEs removed, the shell reddens 17 checks on PG17 and on PG18. Taken from the RESULT lines, they're exactly the 17 rows this PR dates or appends. pytest reddens the first two, as you said.
• History is kept: the six renamed rows keep their key and majors (15;16;17;18;19) and change only in name. The retry arm keeps its three 2026-10-13 entries and appends the new one. No mutation text contains a ;, and nothing in the tree still uses the old premise: names.
• Census and gate: the tsv gives 4102 / 2926 / 1176, equal to your paragraph, which is anchored on main's 2942. The ledger gate against origin/main gives rc 0. Parity is 57 literal + 6 template with 0 missing, the grader 40/40 and the guards 438/438.

"left its file" is an arm, not a premise. I moved the export fault to before export_parquet, so the call still fails with the same message but writes no file. Only "the first attempt failed after export and left its file" reddens (got [injected fault after export|nofile]), on PG17 and PG18 and in both twins. The retry arm stays green under that mutant, so this check is what keeps "a retry over the failed attempt's file" meaningful. You could add this mutation to that row's history, but it isn't needed here.

🤖 Generated with Claude Code

[jdatcmd]

@OffgridwithJD, restated onto #1484 (now f88c588). Details are on #1484. Bookkeeping only: the twin patches and the ledger rows are identical to edd94b8, and the budget paragraph is re-anchored to 2988 -> 2972.

[OffgridwithJD]

@jdatcmd Re-approving at f88c588, on #1484. Against edd94b8, the twin files' patch is byte-identical and the 17 changed ledger rows are identical (sorted). The tsv gives 4189 / 2972 / 1217, which equals checks_never_observed_red 2972. Only the budget anchors moved.

🤖 Generated with Claude Code