Release notes with the version in the filename are history too

src/docproof/config.py

@@ -72,7 +72,24 @@
 # document in the whole corpus and the other two matched none, which is not evidence; and
 # `TOMBSTONE` already draws the line that deprecated is not removed.
 HISTORICAL = re.compile(
-    rf"(?ix) (^|/) (?: {_HISTORICAL_NAMES} ) (?: \.[a-z0-9]+ )? (?: / | $ )"
+    # THE VERSION SUFFIX, which is this rule failing to recognise its own convention rather
+    # than a new class of document. `release[-_ ]?notes` has been on the list above since the
+    # fastapi measurement, and it matched `release-notes.md` while walking straight past
+    # `docs/release_notes_v7.9.0.md` - because after the name it demanded an extension or a
+    # slash, and a version number is neither.
+    #
+    # `enarjord/passivbot`, batch 10. Two of its five findings are `release_notes_v7.9.0.md`
+    # naming `configs/examples/default_trailing_grid_long_npos10.json` and
+    # `release_notes_v7.10.0.md` naming another config, both deleted in later releases. Both
+    # documents are correct: the config existed at the version the file is named for. Pillow
+    # gets this for free because its notes live in a `releasenotes/` DIRECTORY; a project that
+    # keeps them flat and stamps the version on the filename got judged for the same content.
+    #
+    # Measured over every sweep batch, 147 finding lines: this suppresses exactly those two
+    # and nothing else. Deliberately tight - the version must follow immediately, so
+    # `changelog-policy.md` stays judged, and so does anything whose suffix is a word.
+    rf"(?ix) (^|/) (?: {_HISTORICAL_NAMES} )"
+    rf" (?: [-_ ] v? \d+ (?: \.\d+ )* )? (?: \.[a-z0-9]+ )? (?: / | $ )"
     rf" | (^|/) changelog\.d/ "
     rf" | (^|/) archived? / "
     # A FILE called archive, not only a directory. This rule already believes the word means

tests/test_paths.py

@@ -848,3 +848,38 @@ def test_a_plain_deletion_still_reads_as_a_deletion(make_repo: Callable[..., Pat
     assert verdict is Verdict.BROKEN
     assert "deleted in" in detail
     assert "moved to" not in detail
+
+
+def test_release_notes_are_history_even_with_the_version_in_the_name():
+    """`release[-_ ]?notes` has been on `_HISTORICAL_NAMES` since the fastapi measurement,
+    and it could not see `docs/release_notes_v7.9.0.md`: after the name the pattern demanded
+    an extension or a slash, and a version number is neither.
+
+    `enarjord/passivbot` pays for that. Two of its five findings are release notes naming
+    config files deleted in later releases, and both documents are correct - the config
+    existed at the version the filename carries. Pillow escapes this only because its notes
+    sit in a `releasenotes/` DIRECTORY, which the path rule already reads.
+
+    Measured across every sweep batch, 147 finding lines: the widened pattern suppresses
+    exactly those two. The near misses are what keep it honest - a suffix that is a word,
+    not a version, still gets judged.
+    """
+    from docproof.config import is_historical
+
+    for path in (
+        "docs/release_notes_v7.9.0.md",
+        "docs/release_notes_v7.10.0.md",
+        "docs/release-notes-2.rst",
+        "CHANGELOG-1.2.md",
+        "docs/migration_v2.md",
+    ):
+        assert is_historical(path), path
+
+    for path in (
+        "docs/changelog-policy.md",
+        "docs/release-notes-process.md",
+        "docs/history-of-the-parser.md",
+        "docs/newsletter.md",
+        "README.md",
+    ):
+        assert not is_historical(path), path