Two more suites carry the stale not-registered instruction #1087 removed from hilbert_cluster.sh

[OffgridwithJD]

#1087 removed three stale instruction paragraphs from test/hilbert_cluster.sh because they told the next reader to undo work that had already been done. Two sibling files carry the same defect, and one of them is the file #1087's new docs cite as the source of their measurement.

Found by sweeping the class rather than the instance, after #1087 merged at fcfd3e6. Everything below is measured against author/main at that commit.

test/hilbert_locality.sh

# DELIBERATELY NOT REGISTERED in test/run_all_versions.sh yet. Registering it is
# part of the PR that lands it in the matrix; harness_selftest.sh sweeps
# test/*.sh and asserts every suite is registered, so this file makes that arm
# red until then.

Both claims are false on main:

claim	measured
not registered in run_all_versions.sh	it is registered (present in the SUITES array, 253 entries)
"makes that arm red until then"	harness_selftest.sh is green: 967 passed + 0 failed + 0 unrunnable

This is the file docs/features.md, docs/how-to.md and docs/best-practices.md now cite for the 1.24x–2.04x Hilbert-vs-Z-order range (ROWS=200000). So the suite backing alpha4's headline documentation tells a reader it is not in the matrix and is expected to redden the selftest.

test/hilbert_curve.sh

# THIS SUITE IS NOT REGISTERED in test/run_all_versions.sh, so no CI run
# dispatches it. That is deliberate while src/columnar_curve.c does not exist
# and every battery arm is RED by design; registering it belongs in the commit
# that adds the encoder, together with the CHANGELOG entry. Until then
# test/selftest/070 reports it as UNREGISTERED, which is the accurate state.

Three claims, all false:

claim	measured
not registered	it is registered
src/columnar_curve.c does not exist	it exists, 127 lines, and links into pgcolumnar.so
selftest/070 reports it UNREGISTERED, "which is the accurate state"	it is not the accurate state

Why this matters more than a stale comment usually would

Both paragraphs are instructions, in the present tense, and both instruct the reader toward the state the tree has already left. hilbert_curve.sh goes further and tells them registering the suite "belongs in the commit that adds the encoder" — a commit that has landed. #1087's own replacement text put it better than I can:

A stale instruction is worse than a stale fact, because it tells the next person to undo what was done.

Suggested fix

Same shape as #1087's: replace each paragraph with what is true now, and keep whatever part of the original was a real trap worth preserving. Neither file needs a logic change — hilbert_cluster.sh's fix in #1087 was comment-only and these would be too.

Note on how this was found, because the sweep has a trap in it

The forward sweep (files claiming "NOT REGISTERED", checked against the real SUITES list) found both. A reverse sweep I wrote to cross-check — registered suites whose header says the selftest arm is red — found nothing, and that was my instrument failing, not evidence of absence: hilbert_locality.sh wraps the phrase across two comment lines as makes that arm / # red until then, so a line-oriented grep cannot see it. It only worked after joining the comment continuations.

Worth knowing for anyone writing a guard for this class: the claim to match on is the registration status, which is structural, not the prose, which wraps.

test/run_all_versions.sh also matches not registered, but that is its own reconciliation diagnostic ("listed as debt but not registered: $_n") and is correct. Not a finding.

[OffgridwithJD]

Fixed by #1089, merged at cfdb393. Verified on main rather than assuming the merge did what the PR said:

stale not-registered claims, swept with comment continuations joined:
  hilbert_curve      -> only hit is the QUOTED old claim inside its new table
  hilbert_locality   -> no hits at all
  run_all_versions   -> its own reconciliation diagnostic, correct

Both headers now state what is true, and neither instructs a reader to undo work that has landed. The check names did not move (hilbert_curve 38, hilbert_locality 34, identical lists before and after) and both changes were comment-only.

One thing worth leaving here for whoever writes a guard for this class. The sweep that finds it must join comment continuations first: hilbert_locality.sh wrapped the phrase across two lines as makes that arm / # red until then, which a line-oriented grep cannot see, and a reverse check I wrote reported the file clean because of it. Match on registration status against the SUITES array, which is structural, never on the prose, which wraps.

And such a guard will need to exempt quoted text, since the corrected headers deliberately quote the claims they are correcting — the same way selftest/080 exempts the comment that explains the pattern it bans.