docs(contributing): record the tests/ and changelog PR-citation exemptions

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort) — prose only, no Python touched
• make test passes, and a test case covers the change — tests/project/test_conventions_doc_claims.py run directly (the full suite OOMs and is never run)
• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• fix — corrects a defect
• feat — adds a feature
• perf — changes performance, not behaviour
• refactor — changes neither behaviour nor performance
• test — tests only
• docs — documentation only
• ci — workflows or build tooling
• chore — anything else

---

Description of your pull request and other information

Two blocking review findings on docs/source/contributing/conventions/documentation.rst
(the #719 ruling page), both already on main since the page's initial merge — this
corrects a merged page rather than fixing an unmerged one.

1. The "cite the issue, not the pull request" paragraph recorded the ruling without the
exemption the owner granted. A changelog number is the pointer a reader follows, and
tests/** is exempt for the same reason: 21 of its citations, across 12 pull requests,
name one with no issue behind it at all, and the cited fact often lives only in the pull
request's own body or review thread. Both exemptions are now recorded together; the
sibling convention pages remain unconverted as of this change.

2. A sentence on why the root toctrees are :hidden: pointed backward into a grep
code block instead of stating its own reason. Rewritten to state it directly.

Also fixed: two dangling "the first/second pass" references, an unfindable citation of
"the changelog review", and five lines left short of the page's own 88-column wrap.

Verified: tests/project/test_conventions_doc_claims.py — 37 passed, 1 skipped, 140
subtests. No line exceeds 88 columns. Prose only — no role or directive changed.

[JarryShaw]

GOOD TO GO at bf331eecb — sonnet cross-review, first round clean. One soft note taken
anyway, so the head will move once more before this is ready; I will re-confirm on the new sha.

It re-derived every factual claim in the new exemption passage from the tree rather than from the
diff, and all hold. Notably the external-contributor instance: pull request #563 is authored by
Ts-Boom, closes zero issues, and tests/protocols/test_dispatch_default_resolution_unit.py
states exactly that — so "one credits a proposal to an external contributor and closes nothing"
is precise, not rhetorical. Spot-checked the "issue exists but lacks the fact" claim against
tests/corekit/test_sentinels_housing_unit.py: #911 holds the ruling, while the fact that its
__all__ half landed separately lives only in the pull request.

Both blocking defects confirmed fixed. The orphaned demonstrative is gone — :164 now states the
reason instead of pointing at a grep block. "The first pass" has an antecedent. It grepped the
whole page for sentence-initial That/This/It and found the only remaining hits are
pre-existing and untouched here.

It also independently validated the "whole exempt set" claim rather than assuming it. It
suspected pages discussing a pull request as subject matter might deserve a third exemption,
checked the open conversion of the four sibling pages, and found those cases converted too — so
the closed two-member set is correct.

The note I am taking: the passage says the changelog and tests/ are exempt "for the same
reason", but the two sentences after it give two different reasons — the number is the reader's
pointer, versus the cited fact not existing in any issue. The reviewer called it defensible and
not a blocker. I am changing it to "related reasons" regardless: eight rounds on this file have
found exactly one class of defect, a claim slightly stronger than its evidence, and this is one.

One thing worth recording about counts. Its totals disagree with every earlier derivation
again — 358 citations in changelog/1.5.0.rst where I measured 437 under a different definition,
and 23 of 117 pull requests closing no issue. It also tripped on JarryShaw/DictDumper#121/#125
before catching them, which is now three agents and me. That divergence is the argument for having
removed the figures from the page, and they stay out.

UNVERIFIED by it: whether "frequently lacks the fact" holds at scale; it sampled one supporting
instance rather than all ~94.

[JarryShaw]

GOOD TO GO carries to 5a6ca38d0. The diff from the reviewed head bf331eecb is literally
one word on one line — for the same reason → for related reasons, which is precisely what the
review asked for:

1 file changed, 1 insertion(+), 1 deletion(-)

I am carrying a verdict across a head change, which I normally refuse to do, so the reasoning
should be on the record: the only change is the substitution the reviewer itself requested, nothing
else in the diff, and the line is 80 characters either way so the wrap is untouched. Re-verified on
the new head: 0 lines over 88, the page cites only #719, #911, #934, #949 — all issues —
and 37 passed, 1 skipped, 140 subtests. A fresh round here would re-derive the same facts against
an identical page.

Labelled review: good-to-go. Unpublished and awaiting you — I do not merge.

[JarryShaw]

GOOD TO GO at b53819a4c — opus confirming review. Labelled review: good-to-go.

The clause removal is sound and the surviving sentence is true. It read every blob with
git cat-file -p <ref>:<path> rather than the working tree — which is the right instinct, since
I got this wrong an hour ago from a checkout three merges stale — and measured at both
e544a31ca and b53819a4c, naming the shas.

It widened the check beyond what I asked and that is where the value was. Across all 186
.rst files under docs/source/, the only this-repo pull-request citations are those in
changelog/ plus exactly one more — docs/source/changelog.rst:147 (#30) — and that file's own
opening comment calls it "The changelog's index", so it is the changelog rather than a third
exemption. The lookalikes are all non-pull-requests: #106 and #251 are GitHub discussions,
resolved via the GraphQL discussion(number:) query; JarryShaw/DictDumper#125, #121 and
pynetwork/pypcap#116 belong to other repositories. So "the changelog and tests/ are the whole
exempt set" holds across the docs tree, not merely across this directory.

It corrected my brief, and my check as worded would have failed. I told it to confirm the diff
against origin/main is one paragraph:

origin/main..b53819a4c    1 file, +35/-22    <- the whole PR
aca5e19a5..b53819a4c      1 file,  +3/-4     <- the delta since the reviewed revision

Both verified here. The right comparison was against the previously reviewed revision, not main.

One observation it flagged rather than blocked on: now that the sibling conversion has landed, "a
sibling page that cites a pull request is unconverted" describes nothing currently in the tree — it
is a forward-looking rule rather than a description. That is precisely the timeless form the edit was
reaching for, so it stays.

Mechanical: 310 lines, max width exactly 88, 0 over, 0 trailing whitespace; the three new lines
are 85/87/36 columns; citations are #719 ×15, #911, #934, #949 — all issues. Test run under
unittest.TextTestRunner rather than pytest so subTest failures could not be undercounted:
38 run, 0 failures, 0 errors, 1 skipped — the skip pre-existing and unrelated.

Unpublished and awaiting you — I do not merge.