test(accounting): row 6 sees a refusal, a second real corpus, and 34 of 34 mutants

Row 6 was GREEN with R761 100 % rejected: every element of a refused document
is booked as a coded rejection, so u = 0 and d = 0 and the bundle is empty.
`refused_whole` asks that question on its own now -- the build order asked for
an honest red there, and PM re-measured the green on 2026-09-18 with
`okf build` exiting 1 unseen.

A skipped row no longer leaves the verdict unqualified (`GATE GREEN (row 6 not
run: ...)`), and the exit code is non-zero locally when a corpus source is on
the machine and its row did not run. m-2.

N200 Vegbygging:2024 joins R761 as a second real corpus. R761 holds 0 `fig`,
0 formulas and 0 references, so the only real corpus could not have found the
hole in the STS role map; N200 carries 194 citations, 49 figures and 135
footnotes. A `.json` whose root holds an STS node tree is counted as STS
rather than as keys and leaves -- the container is not the content.

M-4: the review's 26 mutants, ported to the code as it stands, plus 8 for the
new checks. 34 of 34 killed. `tools/okf_gate_mutants.py` runs on a copy of the
tree, and a pattern that does not match is an ERROR and exit 2 -- a mutant
that could not be applied was never measured. That fired once, on M13, after a
refactor moved the line it edits.

m-3: `APPROVED_EXCEPTIONS` was read by no row, so approving one changed
nothing. Each pair is now checked against the witness's own vocabulary and the
run says why it moves no denominator. The gate also prints its OWN limits
beside the verdict, m-5 among them.

The product's accounting tests state the new truth instead of the old one:
`okf build --accounting` over the fixture corpus exits 1 with SIX unaccounted
elements in its own vocabulary -- its first real finding, reachable only now
that fixtures carry the constructs. Four shared element names disagree with
the witness, each pinned with its cause; one of the four is a double count
this package makes (a text box's paragraph, once inside the box and again in
the paragraph carrying it).

Three fixture defects were found and fixed while building them, each of which
would have reported a loss the build never had: a shared string table not
related to the workbook, a `graphicData` with no `uri`, and an odt
`styles.xml` without `<office:styles/>`.

Report: docs/2026-09-18-regnskapsgaten-herdet.md

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tests/test_content_accounting.py

@@ -77,7 +77,7 @@ def test_the_log_separates_carried_files_from_rejections(tmp_path: Path) -> None
     code, bundle, _ = _build(CORPUS, tmp_path)
     log = (bundle / "log.md").read_text(encoding="utf-8")
     assert code == 0
-    assert "merged + files carried through a document + coded rejections = 13 + 2 + 0 = 15" in log
+    assert "merged + files carried through a document + coded rejections = 20 + 2 + 0 = 22" in log
     assert "`extractor_unknown`" not in log
 
 
@@ -123,12 +123,75 @@ def _witness(path: Path) -> dict[str, Any]:
     return data
 
 
+#: What the witness counts and this package does not, per file type. Added
+#: 2026-09-18 when the gate's witnesses were widened: rows 2 and 3 of the gate
+#: are RED on exactly these, and that is the finding rather than a regression.
+#: Closing one here turns this test red, which is the point -- the list is the
+#: standing statement of what a bundle built by this package leaves behind.
+NOT_IN_THIS_PACKAGES_VOCABULARY = {
+    "comment",
+    "endnote",
+    "header_footer",
+    "text_box",
+    "note",
+    "hidden_slide",
+    "formula",
+    "hidden_sheet",
+    "annotation",
+    "citation",
+    "math",
+    "figure",
+    "figure_caption",
+}
+
+
+#: Shared element names on which the two now DISAGREE, with the cause of each.
+#: Three are reclassifications -- the witness gives the element its own role
+#: and this package still calls it the ordinary one -- and the fourth is a
+#: double count this package makes. Measured 2026-09-18.
+VOCABULARY_SHIFTS: dict[tuple[str, str], tuple[int, int, str]] = {
+    ("notater-og-skjult.pptx", "slide"): (2, 1, "a hidden slide counts as an ordinary slide"),
+    ("skjult-ark-og-formel.xlsx", "sheet"): (2, 1, "a hidden sheet counts as an ordinary sheet"),
+    ("liste-og-bilde.odt", "paragraph"): (5, 4, "an annotation counts as prose"),
+    (
+        "topptekst-og-kommentar.docx",
+        "paragraph",
+    ): (
+        5,
+        3,
+        "a text box's paragraph is counted TWICE: inside the box, and again in the "
+        "paragraph that carries the box",
+    ),
+}
+
+
 @pytest.mark.parametrize("name", sorted(_witness(FIXTURES / "inventory.json")["documents"]))
-def test_the_inventory_equals_the_witness(name: str) -> None:
+def test_the_inventory_equals_the_witness_on_every_shared_element(name: str) -> None:
+    """The coupling is unchanged where the two vocabularies meet: one number
+    off on a shared element is a red test unless it is in the table above,
+    with its cause written down."""
     pytest.importorskip("pdfplumber")
     want = _witness(FIXTURES / "inventory.json")["documents"][name]["elements"]
-    got = accounting.inventory(CORPUS, CORPUS / name)
-    assert got.counts() == want
+    got = accounting.inventory(CORPUS, CORPUS / name).counts()
+    shared = {k: v for k, v in want.items() if k not in NOT_IN_THIS_PACKAGES_VOCABULARY}
+    differences = {k: (got[k], v) for k, v in shared.items() if got.get(k) != v}
+    expected = {
+        element: (mine, theirs)
+        for (document, element), (mine, theirs, _) in VOCABULARY_SHIFTS.items()
+        if document == name
+    }
+    assert differences == expected
+
+
+@pytest.mark.parametrize("name", sorted(_witness(FIXTURES / "inventory.json")["documents"]))
+def test_the_classes_this_package_does_not_count_are_named(name: str) -> None:
+    """A known-negative: every element the witness counts and this package
+    does not is on the list above, by name. Nothing is missing quietly."""
+    pytest.importorskip("pdfplumber")
+    want = _witness(FIXTURES / "inventory.json")["documents"][name]["elements"]
+    got = accounting.inventory(CORPUS, CORPUS / name).counts()
+    assert set(want) - set(got) <= NOT_IN_THIS_PACKAGES_VOCABULARY
+    assert not set(got) - set(want), "this package counts something no witness does"
 
 
 def test_the_inventory_resolves_the_files_a_document_points_at() -> None:
@@ -144,15 +207,43 @@ def test_the_inventory_of_an_unread_type_is_empty() -> None:
 # --- the door ----------------------------------------------------------------
 
 
-def test_the_door_books_every_element_of_the_fixture_corpus_once(tmp_path: Path) -> None:
+#: The elements this package loses from the fixture corpus, in its OWN
+#: vocabulary, measured 2026-09-18 on the widened corpus. Each is a real loss
+#: with a named cause, and the door exits 1 because of them -- which is the
+#: behaviour the flag was built for, exercised for the first time by fixtures
+#: that actually carry the constructs.
+KNOWN_LOSSES = {
+    ("liste-og-bilde.odt", "image"): (0, 1),
+    ("liste-og-bilde.odt", "paragraph"): (4, 5),
+    ("skjult-ark-og-formel.xlsx", "image"): (0, 1),
+    ("topptekst-og-kommentar.docx", "footnote"): (0, 1),
+    ("topptekst-og-kommentar.docx", "paragraph"): (3, 5),
+}
+
+
+def test_the_door_books_every_element_of_the_fixture_corpus_but_the_known_losses(
+    tmp_path: Path,
+) -> None:
     pytest.importorskip("pdfplumber")
     path = tmp_path / "accounting.json"
     code, _, err = _build(CORPUS, tmp_path, "--accounting", str(path))
-    assert code == 0, err
     data = json.loads(path.read_text(encoding="utf-8"))
     assert data["accounting_version"] == 1
-    assert data["unaccounted"] == 0
     assert data["double_booked"] == 0
+    assert data["unaccounted"] == 6, err
+    assert code == 1, "a run that loses content does not exit zero"
+    found = {}
+    for document in data["documents"]:
+        for kind, number in document["inventory"].items():
+            fate = document.get("fates", {}).get(kind, {})
+            booked = (
+                fate.get("carried", 0)
+                + fate.get("pointer", 0)
+                + sum(fate.get("rejected", {}).values())
+            )
+            if booked != number:
+                found[(document["source_file"], kind)] = (booked, number)
+    assert found == KNOWN_LOSSES
     files = {entry["source_file"]: entry["fate"] for entry in data["files"]}
     assert files == {"graphics/figur-84-1.png": "carried", "graphics/tabell-84-2.png": "carried"}
     web = next(d for d in data["documents"] if d["source_file"] == "prosess-84-web.html")
@@ -163,8 +254,8 @@ def test_the_door_writes_the_accounting_into_the_log(tmp_path: Path) -> None:
     pytest.importorskip("pdfplumber")
     _, bundle, _ = _build(CORPUS, tmp_path, "--accounting", str(tmp_path / "a.json"))
     log = (bundle / "log.md").read_text(encoding="utf-8")
-    assert "* **Accounting**: 13 document(s) and 2 other file(s);" in log
-    assert "0 unaccounted, 0 double-booked." in log
+    assert "* **Accounting**: 20 document(s) and 2 other file(s);" in log
+    assert "6 unaccounted, 0 double-booked." in log
 
 
 def test_a_rejected_document_is_logged_with_what_its_source_held(tmp_path: Path) -> None: