Three last things the corpus was still catching

src/mimem/clean/sections.py

@@ -80,9 +80,19 @@
 
 _LEADING_NUMBER = re.compile(r"^\s*(?:\d+(?:\.\d+)*|[IVXLC]+)[.)]?\s+")
 _EMAIL = re.compile(r"[\w.+-]+@[\w-]+\.[\w.]+")
+#: ``A/S`` is the Norwegian limited-company suffix, and it is spelled in capitals. Written
+#: ``a/?s`` under ``re.I`` alongside everything else it matched the English word **as**, so any
+#: front-matter block containing it was an affiliation -- including one paper's title, "Aluminum
+#: hydride *as* a hydrogen and energy storage material". Triage then dropped the title as
+#: content-free while the orientation went on saying it, which is what rule ``STR-01`` asks for
+#: and what ``COH-01`` then reported as a leak.
 _AFFILIATION_HINT = re.compile(
     r"\b(university|universitet|institute|institutt|department|dept\.|laborator|college|"
-    r"school of|centre|center|academy|hospital|gmbh|inc\.|ltd|a/?s|norway|sweden|denmark)\b",
+    r"school of|centre|center|academy|hospital|gmbh|ltd|(?-i:A/?S)|"
+    # No full stop on these two: the pattern ends in a word boundary, and there is none
+    # between "Dept." and the space after it, so both alternatives never matched anything.
+    r"dept|inc|"
+    r"norway|sweden|denmark)\b",
     re.I,
 )
 #: A reference entry needs *both* a list marker and a bibliographic signal. Requiring only the

src/mimem/verbalize/citations.py

@@ -201,11 +201,16 @@ def _attribute(match: re.Match[str]) -> str:
 #: between them** is not a shape prose has. "Fig. 3" and "Ref. 12" have the space and never
 #: match; a decimal has a digit before its point and never matches either.
 #:
-#: The optional bracket, comma or quotation mark is for a marker that follows a parenthetical
-#: or a quotation: "(LEDC).31", "carbonates),.17" and a simplified "falling cards model".25 all
-#: put something between the last letter and the stop.
+#: The brackets, commas and quotation marks are for a marker that follows a parenthetical or a
+#: quotation: "(LEDC).31", "carbonates),.17" and a simplified "falling cards model".25 all put
+#: something between the last letter and the stop. It is a *run*, because parentheticals nest:
+#: "in N-Methylpyrrolidone (Aldrich)).19" closes two of them at once, and matching a single one
+#: was still one short.
+#:
+#: They are consumed and put back rather than looked behind, because a lookbehind in Python has
+#: to be a fixed width and a run of them is not.
 _SUPERSCRIPT_AFTER_STOP = re.compile(
-    rf"(?:(?<=[A-Za-z][.!?])|(?<=[A-Za-z][)\],\"'”’][.!?]))({_RUN})(?![A-Za-z])"
+    rf"(?P<before>[A-Za-z][)\],\"'”’]*[.!?])(?P<run>{_RUN})(?![A-Za-z])"
 )
 
 #: Below this many markers, the pattern is more likely to be data than a citation style.
@@ -263,7 +268,7 @@ def strip_superscript_citations(text: str) -> str:
     :data:`MIN_SUPERSCRIPT_EVIDENCE`. A single stray digit after a word is far more likely to be
     a typo or a variable than a citation, and deleting it would be silent corruption.
     """
-    text = _SUPERSCRIPT_AFTER_STOP.sub("", text)
+    text = _SUPERSCRIPT_AFTER_STOP.sub(r"\g<before>", text)
     return _SUPERSCRIPT_AFTER_WORD.sub(
         lambda m: (
             m.group(0) if _is_unit_exponent(text, m) or _is_formula_subscript(text, m) else ""

src/mimem/verbalize/numbers.py

@@ -258,7 +258,13 @@ def __init__(
                 # at all and "1.5" reached the audio as a bare number. The plain pass could not
                 # take it either: it will not start on a digit that follows a letter.
                 re.compile(
-                    r"(?<![\w.])(?P<token>(?=\w*[A-Za-z])(?=\w*\d)[A-Za-z0-9]+(?:\.\d+)?)"
+                    # ...or straight after a number, when a letter starts the token. A hydrate
+                    # is written that way -- "AlH3\u00b70.25Et2O" -- and the lookbehind refused
+                    # to begin on "Et2O" because a digit was in front of it, so the plain pass
+                    # said "zero point two five" and left the ether spelled out as a number.
+                    # The leftmost match still wins, so "4AlH3" is claimed whole as before.
+                    r"(?:(?<![\w.])|(?<=\d)(?=[A-Za-z]))"
+                    r"(?P<token>(?=\w*[A-Za-z])(?=\w*\d)[A-Za-z0-9]+(?:\.\d+)?)"
                     r"(?!\w)(?!\.\d)"
                 ),
                 "designation",

tests/unit/test_clean_text.py

@@ -220,3 +220,30 @@ def test_only_headings_are_split() -> None:
         ],
     )
     assert len(split_stacked_headings(doc).blocks) == 1
+
+
+def test_the_norwegian_company_suffix_is_not_the_english_word_as() -> None:
+    """``a/?s`` under ``re.I``, alongside "university" and "institute", matched **as**.
+
+    Any front-matter block containing that word was called an affiliation -- including one
+    paper's title, "Aluminum hydride *as* a hydrogen and energy storage material". Triage then
+    dropped the title as content-free while the orientation went on saying it, which is what
+    rule STR-01 asks for and what COH-01 then reported as a leak.
+    """
+    from mimem.clean.sections import _AFFILIATION_HINT
+
+    assert not _AFFILIATION_HINT.search("Aluminum hydride as a hydrogen storage material")
+    assert not _AFFILIATION_HINT.search("the capacity as measured after cycling")
+    assert _AFFILIATION_HINT.search("Elkem AS, Kristiansand")
+    assert _AFFILIATION_HINT.search("Institutt for energiteknikk, Kjeller")
+
+
+def test_two_hints_that_a_full_stop_had_made_unreachable() -> None:
+    """The pattern ends in a word boundary, and there is none between "Dept." and the space."""
+    from mimem.clean.sections import _AFFILIATION_HINT
+
+    assert _AFFILIATION_HINT.search("Dept. of Chemistry, Uppsala")
+    assert _AFFILIATION_HINT.search("Elkem Inc., Pittsburgh")
+    # ...without swallowing the ordinary words they are prefixes of.
+    assert not _AFFILIATION_HINT.search("we incorporated the binder")
+    assert not _AFFILIATION_HINT.search("the incident light was filtered")

tests/unit/test_corpus_defects.py

@@ -506,3 +506,39 @@ def test_a_marker_after_a_quotation_mark() -> None:
 )
 def test_a_designation_may_have_a_decimal_point_in_its_name(written: str, spoken: str) -> None:
     assert _spoken(written) == spoken
+
+
+# -- the last three ----------------------------------------------------------------------------
+
+
+def test_a_marker_after_more_than_one_closing_bracket() -> None:
+    """Parentheticals nest, and matching a single closing bracket was still one short.
+
+    "...5 wt% of Super P carbon (Tim-cal) in N-Methylpyrrolidone (Aldrich)).19 Electrochemical
+    tests were performed..." -- two brackets close at once between the sentence's last letter
+    and its full stop.
+    """
+    written = "in N-Methylpyrrolidone (Aldrich)).19 Electrochemical tests were performed"
+    assert strip_superscript_citations(written) == (
+        "in N-Methylpyrrolidone (Aldrich)). Electrochemical tests were performed"
+    )
+
+
+@pytest.mark.parametrize(
+    ("written", "spoken"),
+    [
+        # A hydrate is written with the water or the ether welded to the stoichiometry, and the
+        # designation pattern refused to begin on a token with a digit in front of it -- so the
+        # plain pass said "zero point two five" and left the ether spelled out as a number.
+        (
+            "a composition close to AlH3·0.25Et2O and although it is amorphous",
+            "a composition close to AlH three times zero point two five Et two O "
+            "and although it is amorphous",
+        ),
+        # The leftmost match still wins, so a coefficient in front is claimed with the token.
+        ("4AlH3 was distilled off", "four AlH three was distilled off"),
+        ("3LiH and AlCl3 react", "three LiH and AlCl three react"),
+    ],
+)
+def test_a_designation_may_begin_where_a_number_ends(written: str, spoken: str) -> None:
+    assert _spoken(written) == spoken