docs(tests): stop calling pull requests issues in test prose

[JarryShaw]

Please follow the guide below

• Searched for similar pull requests
• Followed the coding style
• make test passes, and a test case covers the change — prose only, no assertion or logic touched
• Added a changelog entry — 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?

• docs — documentation only

---

Description of your pull request and other information

Accuracy sweep for #719: nine places in tests/ call a pull request an issue. Each number was resolved with gh api rather than guessed, and the house form already used elsewhere in tests/ — "GitHub pull request #N" — is what replaces it.

File:line	#	really	said
foundation/registry/test_protocols.py:535	815	PR	issue
corekit/test_enum_lookup_reparent_930_unit.py:5	921	PR	issue
corekit/test_fields_numbers_port_option_no_mint_unit.py:235	764	PR	issue
corekit/test_sentinel_exports_unit.py:237	906	PR	issue
project/test_conventions_doc_claims.py:805	936	PR	issue
protocols/schema/test_enum_schema_registry_unit.py:383	721	PR	issue
test_docstring_contract.py:4	501	PR	issue
utilities/test_exceptions_excepthook.py:440	983	PR	issue
protocols/test_dispatch_default_resolution_unit.py:8	425 / 428	issue / PR	"issues #425/#428"

The sweep extracted 2,190 #NNN sites over 353 distinct numbers by tokenize token kind, resolved all 350 numbers at or above 300 (230 issues, 120 pull requests), and read every flagged line before changing it. Numbers below 300 are packet-diagram labels, counts and widths, and were left alone.

Two sites that look wrong and are not, left untouched: test_final_enforcement.py:121 describes a pull request that implemented issue #778, and const/test_const_ethertype_862_unit.py:212 describes PR #865's review of issue #862's fix. Both are accurate as written.

Citation markup is unchanged — tests/ is deliberately outside #989's :issue:/:pr: conversion, and GH-nnn is untouched.

Prose only: ast.dump with every docstring blanked is identical to main in all 9 files, and the diff is 9 lines for 9 lines. The 9 modules run together give 155 passed / 1 skipped / 340 subtests / 0 failures on this branch and byte-for-byte the same on main.

[JarryShaw]

NEEDS CHANGES at 19cd484d3 — opus cross-review, a different model from the sonnet agent that authored the diff. Eight of the nine fixes are right and well-evidenced. One introduced a worse error than it removed, and separately I have a correction of my own to make.

The defect — tests/protocols/test_dispatch_default_resolution_unit.py:8. It now reads "the defect GitHub issue #425 and pull request #428 fixed at this layer". An issue does not fix a defect; it reports one. #425 is the bug report ("Option, chunk and block registries leak on lookup miss"), and #428's body opens by closing it. The original issues #425/#428 fixed was a single compound reference whose only error was of kind — splitting the kinds apart dropped #425 into the sentence's own "X fixed at layer L" template and made it assert something false. Being fixed: …the defect GitHub issue #425 reported and pull request #428 fixed at this layer….

Correcting myself: I attributed five identity failures to a pytest-randomly ordering artefact. That plugin is not installed. Measured just now — randomly, pytest_randomly, xdist and pytest_subtests are all absent from the venv, -p randomly dies with No module named 'randomly', and pyproject.toml sets no addopts. So -p no:randomly pinned nothing and my mechanism was fiction. What survives is the part that was measured: 155 passed / 1 skipped / 340 subtests / 0 failures, identical on 326a0e5ae and 19cd484d3, and the branch cannot cause those failures because it is docstring-and-comment-only — ast.dump identical in all 9 files and all 1,176 non-docstring string constants identical in order. The order-dependence itself is already tracked by #981 and #674.

The review also strengthened the headline case: #815's body defers to #801, but the maintainer's ruling is in #815's own comment thread and was implemented at e683be072 — so the ruling does live on the pull request and the fix is right. The two untouched sites are confirmed correct: #788 is the PR that implemented issue #778, and #865 is the PR that reviewed issue #862's fix.

Coverage, by a method independent of mine: a raw-text scan over all 224 test modules found 598 kind-adjacent references and no other wrong one, and no number below 349 carries an adjacent kind word, so the sub-300 cutoff hid nothing.

[JarryShaw]

Fixed at 141d4debf — one line, and I verified it is the only change from 19cd484d3.

-defect GitHub issue #425 and pull request #428 fixed at this layer and #560 fixed at the schema
+defect GitHub issue #425 reported and pull request #428 fixed at this layer, and #560 fixed at the schema

#560 confirmed a pull request, so "#560 fixed at the schema layer" is accurate as it stands. Still 9 files, 9 insertions, 9 deletions, one commit on 326a0e5ae, ast.dump identical to main in all 9 files with the non-docstring string constants matching in order, and the nine modules together give 155 passed / 1 skipped / 340 subtests / 0 failures.

While checking the surrounding context I resolved the module docstring's own opening citation: #574 is an issue ("Memoise the default ModuleDescriptor on a registry miss"), so "Regression tests for GitHub issue #574" on line 1 is already correct and stays.

Label back to review: pending — the head moved, so the earlier verdict no longer tracks it. A delta review is running.

[JarryShaw]

GOOD TO GO at 141d4debf — opus delta round. The corrected sentence is true and unambiguous, and the review established the part I could not: #560 really did fix the schema layer — its body adds an _EnumRegistry whose __missing__ returns the default without writing, and #428's own "deliberately not fixed" list names that schema-side follow-up as out of its scope. So all three roles hold: #425 reported, #428 fixed this layer, #560 fixed the schema layer.

On the wording: "pull request #428" is an explicit new subject between "reported" and "fixed", so "reported" cannot bind to both — the only available parse is the intended one.

Also confirmed: delta is one line, whole change still 9 files / 9 / 9, ast.dump identical to main in all nine with non-docstring string constants matching in order and count, nine modules together 155 passed / 1 skipped / 340 subtests / 0 failures, #574 on line 1 is an issue, and line 13's "pull request #563" is correct — #563 is a PR.

Two non-blocking items, and I am correcting one of them before anyone relies on it.

• Line 8 is now 105 characters (79 on main, 95 before the amend), the longest added line in the commit; the other eight land at 73–92. But the review called it the file's lone prose outlier against two unwrappable :meth: lines — measured, 13 lines in this file already exceed 80: 5, 8, 24, 28, 34, 54, 122, 131, 185, 186, 187, 192, 194. So the file is not strictly 80-wrapped, nothing lints line length here (no setup.cfg, .flake8, tox.ini or .pylintrc, and pyproject.toml sets none), and reST reflows the paragraph. Not worth a CI round to rewrap, which is a judgement rather than a measurement — say so if you disagree.
• A lineage point that predates this PR and is out of its scope: Option, chunk and block registries leak on lookup miss, the same way __proto__ did #425 reported the __option__/__chunk__/__block__ family, while the __proto__ pollution this file actually tests was reported by Parsing a packet pollutes the shared __proto__ registry, so a later register_* warns about a protocol nobody registered #421 and fixed by protocols: stop next-layer dispatch writing to shared registries, and fix two transport defects #426. protocols: stop option, chunk and block dispatch writing to shared registries #428 does touch this layer — it added the guard that stops a default-factory descriptor being memoised, which is exactly what these tests pin — so "fixed at this layer" is sound. "The defect issue Option, chunk and block registries leak on lookup miss, the same way __proto__ did #425 reported" is loose for this layer. Recording it on docs: sweep all prose for concision and accuracy, and drop timed context outside the changelog #719 rather than widening this change.

[JarryShaw]

Ready to merge at 141d4debf — CI complete: 69 CheckRun legs green, 3 skipped, 0 failures, 0 in flight, mergeStateStatus CLEAN. The opus delta verdict stands at this head; nothing has been pushed since.

Yours to merge. The two items I left out are recorded rather than dropped: line 8 is 105 characters in a file that already has 12 other lines over 80 and no line-length linter, and the #425-versus-#421 lineage question is noted on #719 for the next pass over that file.