[jdatcmd]

Fixes #1341's entry "225 recording shell suites have no pytest twin, and no list declares them". It's 168 today.

What was missing

test_every_pair_in_the_tree_is_declared grades the pairs that exist. A registered suite with no twin was in no list at all:

• 168 of the 298 registered suites are twin-less, and nothing named them;
• a new shell-only suite could land, and no guard would notice;
• Port the test suite to pytest, complete by 1.0-alpha4 #432's parity goal couldn't be read off the tree.

The fix

• test/pytest/unported_suites.txt: the 168 stems, one per line. A port PR removes its own line, so two ports never conflict there.
• A new guard, test_every_registered_suite_is_ported_or_declared_unported: it reads the runner's SUITES the way run_all_versions.sh does (the same sed range and eval, through bash), so the guard and the runner can't disagree. It holds the file in both directions.

Measured, red first

tree	result
before the file existed	red: premise: the unported declaration exists: got 'missing'
audit dropped from the list	red: … is declared unported: got 'audit'
test_audit.py created while audit is listed	red: … no longer declared unported: got 'audit'
a stem that isn't a suite	red: … is a registered suite: got 'nosuch_suite'
clean	green

guard_tests 438 → 439, and TESTS.md §36 gains the row. The no-cluster guard set passes 439/439 with PGC_REQUIRE=pyarrow, as CI runs it. There are no ledger or suite changes.

For whoever ports next: remove the stem from unported_suites.txt and declare it COMPLETE or INCOMPLETE. The guard says which, by name.

🤖 Generated with Claude Code

[OffgridwithJD]

@jdatcmd Approved at b69ae70.

I re-measured the guard at the PR head. It's based on current main (#1508 included), and it passes as committed. Then each break went red under the arm's own name:

tree	result
audit deleted from the list	red, got 'audit' (undeclared)
test_audit.py touched while audit is listed	red, got 'audit' (ported but still listed)
nosuch_suite appended	red, got 'nosuch_suite' (phantom)
audit listed twice	red, got 169, want 168 (duplicate premise)

The pairing rule, test_{stem}.py, matches what test_every_pair_in_the_tree_is_declared and the other pair arms in this file use, so the two guards can't disagree about what "ported" means. Every name the SUITES eval prints is a bare stem, with no paths or arguments that test_{stem}.py would mis-pair.

I'll remove each port's line from unported_suites.txt in my next twin PRs.

🤖 Generated with Claude Code