[JarryShaw]

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Changelog entry: N/A -- reverts an unreleased, undocumented docs: #1103 round 10 -- close the whole-repo prose sweep #2432 change back to the prior, already-ratified behavior; net effect on 1.5.0 is zero

What is the purpose of your pull request?

• fix -- corrects a defect

---

Description

#2432 (round 10 of #1103) mistakenly changed register's RegistryError message
from "...must be a Protocol subclass" to "...must be a ProtocolBase subclass"
in pcapkit/protocols/protocol.py and foundation/registry/protocols.py, and
updated the three locked assertions in tests/protocols/test_register_class_guard_unit.py
(two) and tests/foundation/registry/test_protocols.py (one) to match.

Naming Protocol (not ProtocolBase) in that message is a deliberate, ratified
decision, not stale prose: #1016 ruled the message's reader is third-party code
that should be told to subclass Protocol, since ProtocolBase's own docstring
says it is "for internal use only". #2316 re-confirmed this is out of scope to
change. #2420's merge commit records the split directly: "The Raises: clause
of each register keeps naming ProtocolBase, unchanged... the raised message
names Protocol." git log -S on both wordings shows #2432 is the only commit
that ever flipped the message; #2420 only touched the Raises: docstring
clause, never the message. This PR reverts both raise sites and the three
locked test assertions back to Protocol subclass.

Also fixes a citation #2432 left behind: tests/foundation/reassembly/ test_deferred_pickle_2423_unit.py cited :issue:2424 (the merged PR, not an issue) where the sentence means the underlying fix, tracked as #2423 -- now `:issue:`2423, matching the file's own existing citation style and the
changelog's (:issue:2423, :issue:2425, :pr:2424).

#2432's other two fixes (a stale PR-history docstring, a cross-reference to a
moved EnumField) were correct and are untouched.

Refs #1016, #2316, #2420, #1103

[JarryShaw]

Review: GOOD TO GO at dfda0f0a2 (Opus, different model from the author). The code is a correct, complete, byte-exact revert of #2432's mistake, verified independently — but the review found the PR description carried three factual errors, now corrected in the body above (this becomes the squash commit message, same as #2432):

• Said "four" locked assertions; there are three (two in test_register_class_guard_unit.py, one in test_protocols.py).
• Misquoted docs: prefer Protocol over ProtocolBase in cross-references, per the owner's #1103 ruling #2420's commit text — fixed to the actual wording.
• Falsely claimed docs: prefer Protocol over ProtocolBase in cross-references, per the owner's #1103 ruling #2420 had already fixed a prior accidental flip of this exact message; it hadn't — git log -S on both wordings confirms docs: #1103 round 10 -- close the whole-repo prose sweep #2432 is the only commit that ever flipped the message. docs: prefer Protocol over ProtocolBase in cross-references, per the owner's #1103 ruling #2420 only touched the unrelated Raises: docstring clause.

Independently confirmed: the ruling itself (#1016/#2316, with the caveat that it's one shared login throughout), the diff has no scope creep (5 files, 6 lines each way, byte-exact against pre-#2432), the citation fix is correct, the raised message text is restored (register_protocol(object) → "must be a Protocol subclass"), and re-breaking the message confirms the tests actually catch the regression (17 failures). Full suite / lint / type checks not run (budget) — reported as unverified rather than assumed clean.