feat(#5): шаг 2 — числа правее последней подписанной колонки не колонки сумм

Столбец без подписи правее последней колонки с подписью периода колонкой сумм
не считается; вместо COLUMN_WITHOUT_DATE — info NUMBERS_OUTSIDE_TABLE с адресами
ячеек. Решение владельца 18.10.2026 (arm-ka-docs#116, Р-31).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

README.md

@@ -33,7 +33,7 @@ make venv && make check
 | `DocStatus` | `DONE = "done"`, `NEEDS_CHECK = "needs_check"` |
 | `EnvelopeStatus` | `DONE = "done"`, `NEEDS_CHECK = "needs_check"`, `ERROR = "error"` |
 | `SignForm` | `PARENTHESES = "parentheses"`, `MINUS = "minus"`, `UNSIGNED = "unsigned"` |
-| `WarningCode` | по члену на код словаря `statement_warnings.yaml`, имя = значение: `JSON_FAILED`, `TABLE_PARSE_ERROR`, `TABLE_LOST`, `TABLE_SKIPPED`, `CODE_COLUMN_NOT_FOUND`, `SEGMENT_WITHOUT_CODES`, `SHEET_NOT_JOINED`, `CONTINUATION_TYPE_CONFLICT`, `SEGMENT_COLUMNS_MISMATCH`, `COLUMN_WITHOUT_DATE`, `COLUMN_DATE_UNCERTAIN`, `HEADER_ROW_NOT_FOUND`, `TYPE_CONFLICT`, `NO_ORGANIZATION_NO_INN`, `PERIOD_UNKNOWN`, `REPORT_DATE_UNCERTAIN`, `XLSB_SCALE_SUSPECT`, `INN_NOT_FOUND`, `INN_UNCERTAIN`, `ORGANIZATION_NOT_FOUND`, `ORGANIZATION_UNCERTAIN`, `REPORT_DATE_UNPARSED`, `UNITS_NOT_FOUND`, `FILE_TYPE_MISMATCH`, `COLUMN_SKIPPED`, `FRACTIONAL_IN_THOUSANDS`, `XLSB_NO_FORMATS`, `NUMBER_FORMAT_UNPARSED`, `VALUE_OUT_OF_RANGE`, `SHEET_OTHER`, `ROW_SPLIT`, `XLS_NO_CODEPAGE`. Коды `OSV_*` пути ОСВ сюда не входят (их создаёт tabular-parser, спека «Шаг 8 — предупреждения») |
+| `WarningCode` | по члену на код словаря `statement_warnings.yaml`, имя = значение: `JSON_FAILED`, `TABLE_PARSE_ERROR`, `TABLE_LOST`, `TABLE_SKIPPED`, `CODE_COLUMN_NOT_FOUND`, `SEGMENT_WITHOUT_CODES`, `SHEET_NOT_JOINED`, `CONTINUATION_TYPE_CONFLICT`, `SEGMENT_COLUMNS_MISMATCH`, `COLUMN_WITHOUT_DATE`, `COLUMN_DATE_UNCERTAIN`, `HEADER_ROW_NOT_FOUND`, `TYPE_CONFLICT`, `NO_ORGANIZATION_NO_INN`, `PERIOD_UNKNOWN`, `REPORT_DATE_UNCERTAIN`, `XLSB_SCALE_SUSPECT`, `INN_NOT_FOUND`, `INN_UNCERTAIN`, `ORGANIZATION_NOT_FOUND`, `ORGANIZATION_UNCERTAIN`, `REPORT_DATE_UNPARSED`, `UNITS_NOT_FOUND`, `FILE_TYPE_MISMATCH`, `COLUMN_SKIPPED`, `FRACTIONAL_IN_THOUSANDS`, `XLSB_NO_FORMATS`, `NUMBER_FORMAT_UNPARSED`, `VALUE_OUT_OF_RANGE`, `SHEET_OTHER`, `ROW_SPLIT`, `NUMBERS_OUTSIDE_TABLE`, `XLS_NO_CODEPAGE`. Коды `OSV_*` пути ОСВ сюда не входят (их создаёт tabular-parser, спека «Шаг 8 — предупреждения») |
 
 ### Вход шага 1 (строит адаптер хоста)
 

arm_statements/columns.py

@@ -14,6 +14,7 @@ from arm_statements.loader import Rules
 from arm_statements.model import Cell, PeriodKind, Source, StatementColumn, StatementHeader, StatementWarning
 
 _FULL_CONFIDENCE = 100.0
+_MAX_REFS = 10
 
 
 @dataclass(frozen=True, slots=True)
@@ -94,7 +95,10 @@ def segment_columns(  # noqa: PLR0913, PLR0917 - сигнатура задана
     """Колонки значений сегмента и предупреждения о пропущенных/недатированных столбцах."""
     drafts: list[ColumnDraft] = []
     warnings: list[StatementWarning] = []
-    for col in range(code_col + 1, _width(rows, [*header_rows, *data_rows])):
+    columns = range(code_col + 1, _width(rows, [*header_rows, *data_rows]))
+    last_dated = _last_dated_column(rows, header_rows, columns, rules)
+    outside: list[int] = []
+    for col in columns:
         numbers = has_numbers(rows, data_rows, col, code_col, source, rules)
         if not header_rows:
             if numbers:
@@ -107,14 +111,62 @@ def segment_columns(  # noqa: PLR0913, PLR0917 - сигнатура задана
             warnings.append(make_warning(WarningCode.COLUMN_SKIPPED, message, rules, column_index=col))
         elif cell is not None and parsed.kind is not PeriodKind.NONE:
             drafts.append(ColumnDraft(col, cell.text, cell.confidence, parsed.kind, parsed.date))
+        elif numbers and cell is None and last_dated is not None and col > last_dated:
+            outside.append(col)  # числа правее последней колонки с подписью периода — не колонка сумм
         elif numbers:
             label, conf = (cell.text, cell.confidence) if cell is not None else ("", None)
             message = f"у столбца {col} с числами нет разобранной подписи периода"
             warnings.append(make_warning(WarningCode.COLUMN_WITHOUT_DATE, message, rules, column_index=col))
             drafts.append(ColumnDraft(col, label, conf, PeriodKind.NONE, None))
+    if outside:
+        warnings.append(_outside_table_warning(rows, data_rows, outside, code_col, source, rules))
     return drafts, warnings
 
 
+def _last_dated_column(rows: Grid, header_rows: tuple[int, ...], columns: range, rules: Rules) -> int | None:
+    """Правый столбец сегмента с подписью периода (дата или относительная подпись); без шапки — `None`."""
+    dated = [
+        col
+        for col in columns
+        if header_rows
+        and (cell := _label(rows, header_rows, col)) is not None
+        and not (parsed := parse_period_label(cell, rules)).excluded
+        and parsed.kind is not PeriodKind.NONE
+    ]
+    return max(dated, default=None)
+
+
+def _outside_table_warning(  # noqa: PLR0913, PLR0917 - внутренняя сборка предупреждения
+    rows: Grid, data_rows: Sequence[int], outside: Sequence[int], code_col: int, source: Source, rules: Rules
+) -> StatementWarning:
+    refs = [
+        cell.ref
+        for r in _code_rows(rows, data_rows, code_col, rules)
+        for col in outside
+        if (cell := cell_at(rows, r, col)) is not None and _is_number(cell, source, rules)
+    ]
+    return outside_table_warning(refs, rules)
+
+
+def outside_table_warning(refs: Sequence[str], rules: Rules) -> StatementWarning:
+    """`NUMBERS_OUTSIDE_TABLE` (info): адреса ячеек правее таблицы формы (в `locator` все, в тексте первые)."""
+    shown = ", ".join(refs[:_MAX_REFS]) + (f" и ещё {len(refs) - _MAX_REFS}" if len(refs) > _MAX_REFS else "")
+    message = f"числа правее таблицы формы не учитываются: {shown}"
+    return make_warning(WarningCode.NUMBERS_OUTSIDE_TABLE, message, rules, locator=", ".join(refs))
+
+
+def merge_outside_table(warnings: Sequence[StatementWarning], rules: Rules) -> list[StatementWarning]:
+    """Предупреждения `NUMBERS_OUTSIDE_TABLE` сегментов — в одно (адреса без повторов)."""
+    refs: dict[str, None] = {}
+    rest: list[StatementWarning] = []
+    for warning in warnings:
+        if warning.code == WarningCode.NUMBERS_OUTSIDE_TABLE:
+            refs.update(dict.fromkeys((warning.locator or "").split(", ")))
+        else:
+            rest.append(warning)
+    return [*rest, outside_table_warning(list(refs), rules)] if refs else rest
+
+
 def _latest(drafts: Sequence[ColumnDraft], *, confident: bool) -> date | None:
     days = [
         d.date

arm_statements/issues.py

@@ -45,6 +45,7 @@ class WarningCode(StrEnum):
     VALUE_OUT_OF_RANGE = "VALUE_OUT_OF_RANGE"
     SHEET_OTHER = "SHEET_OTHER"
     ROW_SPLIT = "ROW_SPLIT"
+    NUMBERS_OUTSIDE_TABLE = "NUMBERS_OUTSIDE_TABLE"
 
 
 def make_warning(  # noqa: PLR0913 -- сигнатура задана публичным API пакета

arm_statements/rules/statement_warnings.yaml

@@ -34,3 +34,4 @@ warnings:
   VALUE_OUT_OF_RANGE: {severity: info, scope: document}
   SHEET_OTHER: {severity: info, scope: file}
   ROW_SPLIT: {severity: info, scope: document}
+  NUMBERS_OUTSIDE_TABLE: {severity: info, scope: document}

arm_statements/segments.py

@@ -5,7 +5,14 @@ from dataclasses import dataclass
 from typing import NamedTuple
 
 from arm_statements.codes import code_parts, is_code_cell, majority_section, parse_code
-from arm_statements.columns import ColumnDraft, DocumentDates, document_dates, finalize_columns, segment_columns
+from arm_statements.columns import (
+    ColumnDraft,
+    DocumentDates,
+    document_dates,
+    finalize_columns,
+    merge_outside_table,
+    segment_columns,
+)
 from arm_statements.issues import WarningCode, make_warning
 from arm_statements.layout import (
     Grid,
@@ -223,7 +230,18 @@ def find_layout(candidate: FormCandidate, header: StatementHeader, rules: Rules)
     if defining is not None:
         dates = document_dates(defining.drafts, header)
         columns, finalize_warnings = finalize_columns(defining.drafts, dates, rules)
-        warnings.extend(defining.draft_warnings)
+        # числа правее таблицы — по всем сегментам (формула аналитика может стоять на любой странице формы)
+        warnings.extend(
+            merge_outside_table(
+                [
+                    w
+                    for seg in segments
+                    for w in seg.draft_warnings
+                    if seg is defining or w.code == WarningCode.NUMBERS_OUTSIDE_TABLE
+                ],
+                rules,
+            )
+        )
         warnings.extend(finalize_warnings)
         if not any(s.header_rows for s in coded):
             warnings.append(

arm_statements/tests/test_columns.py

@@ -68,18 +68,61 @@ def test_junk_label_over_numbers_is_kept(rules: Rules) -> None:
     assert [(w.code, w.severity, w.column_index) for w in warnings] == [("COLUMN_WITHOUT_DATE", Severity.ERROR, 3)]
 
 
-def test_empty_label_over_numbers_is_kept(rules: Rules) -> None:
+def test_empty_label_right_of_dated_columns_is_outside_table(rules: Rules) -> None:
     rows: list[list[Any]] = [
         ["Наименование", "Код", "На 30.09.2025", ""],
         ["Запасы", "1210", "5", "4"],
         ["Итого", "1200", "5", "4"],
     ]
     columns, warnings = run(rows, OCR, rules)
-    assert [(c.column_index, c.label, c.label_confidence, c.period_kind) for c in columns] == [
-        (2, "На 30.09.2025", 100.0, PeriodKind.ABSOLUTE),
-        (3, "", None, PeriodKind.NONE),
+    assert [(c.column_index, c.period_kind) for c in columns] == [(2, PeriodKind.ABSOLUTE)]
+    assert [(w.code, w.severity, w.column_index) for w in warnings] == [("NUMBERS_OUTSIDE_TABLE", Severity.INFO, None)]
+    assert warnings[0].locator == "row:1, row:2"
+
+
+def test_numbers_right_of_last_dated_column_are_not_amounts(rules: Rules) -> None:
+    rows: list[list[Any]] = [
+        ["Наименование", "", "Код", "На 30.09.2025", "На 31.12.2024", None, None],
+        ["Запасы", "", "1210", D("5"), D("4"), None, D("9")],
+        ["Баланс", "", "1600", D("5"), D("4"), D("7"), None],
+        ["Итого", "", "1700", D("5"), D("4"), None, None],
     ]
-    assert [(w.code, w.column_index) for w in warnings] == [("COLUMN_WITHOUT_DATE", 3)]
+    columns, warnings = run(rows, EXCEL, rules, data_rows=(1, 2, 3))
+    assert [c.column_index for c in columns] == [3, 4]
+    assert [(w.code, w.severity) for w in warnings] == [("NUMBERS_OUTSIDE_TABLE", Severity.INFO)]
+    assert warnings[0].locator == "Лист1!G2, Лист1!F3"
+    assert "G2" in warnings[0].message and "F3" in warnings[0].message
+
+
+def test_numbers_outside_table_do_not_depend_on_cell_count(rules: Rules) -> None:
+    rows: list[list[Any]] = [["Наименование", "", "Код", "На 30.09.2025", None]]
+    rows += [["Строка", "", str(1200 + i), D("1"), D("2")] for i in range(12)]
+    cand = excel([("Лист1", rows)])
+    _, warnings = segment_columns(cand.rows, (0,), tuple(range(1, 13)), 2, EXCEL, rules)
+    assert len(warnings) == 1
+    assert warnings[0].message.endswith("и ещё 2")
+
+
+def test_no_dated_columns_keeps_old_behaviour(rules: Rules) -> None:
+    rows: list[list[Any]] = [
+        ["Наименование", "", "Код", "", ""],
+        ["Запасы", "", "1210", D("5"), D("4")],
+        ["Итого", "", "1200", D("5"), D("4")],
+    ]
+    columns, warnings = run(rows, EXCEL, rules)
+    assert [c.column_index for c in columns] == [3, 4]
+    assert [(w.code, w.column_index) for w in warnings] == [("COLUMN_WITHOUT_DATE", 3), ("COLUMN_WITHOUT_DATE", 4)]
+
+
+def test_undated_column_between_dated_keeps_old_behaviour(rules: Rules) -> None:
+    rows: list[list[Any]] = [
+        ["Наименование", "", "Код", "На 30.09.2025", None, "На 31.12.2024"],
+        ["Запасы", "", "1210", D("5"), D("9"), D("4")],
+        ["Итого", "", "1200", D("5"), D("9"), D("4")],
+    ]
+    columns, warnings = run(rows, EXCEL, rules)
+    assert [c.column_index for c in columns] == [3, 4, 5]
+    assert [(w.code, w.severity, w.column_index) for w in warnings] == [("COLUMN_WITHOUT_DATE", Severity.ERROR, 4)]
 
 
 def test_two_row_header_is_joined_with_min_confidence(rules: Rules) -> None:

arm_statements/tests/test_loader.py

@@ -43,6 +43,7 @@ SPEC_WARNINGS = {
     "VALUE_OUT_OF_RANGE": ("info", "document"),
     "SHEET_OTHER": ("info", "file"),
     "ROW_SPLIT": ("info", "document"),
+    "NUMBERS_OUTSIDE_TABLE": ("info", "document"),
     "XLS_NO_CODEPAGE": ("info", "file"),
 }
 

arm_statements/tests/test_pipeline.py

@@ -194,3 +194,19 @@ def test_document_warnings_order(rules: Rules) -> None:
     ]
     assert [c for c in codes if c in expected] == [c for c in expected if c in codes]
     assert codes[0] == "NUMBER_FORMAT_UNPARSED" and codes.index("TYPE_CONFLICT") < codes.index("NO_ORGANIZATION_NO_INN")
+
+
+# --- числа правее таблицы формы (решение владельца 18.10.2026): не колонки сумм, статус не портят ---
+
+
+def test_numbers_right_of_table_do_not_break_status_in_repeated_segments(rules: Rules) -> None:
+    rows: list[list[str | None]] = [[*r, None] for r in bal_rows()]
+    rows[6][3] = "7"  # формула аналитика правее таблицы, в строке БАЛАНС (1600)
+    rows[9][3] = "8"
+    rows.insert(7, [*HEAD, None])  # повторная шапка второго сегмента (страница 2 формы)
+    cand = ocr(rows, header=FULL_HEADER, type_hint=DocumentType.BALANCE)
+    doc = normalize_candidate(cand, rules).document
+    assert doc is not None and doc.status is DocStatus.DONE
+    assert [c.column_index for c in doc.columns] == [2]
+    assert [(w.code, w.severity) for w in doc.warnings] == [("NUMBERS_OUTSIDE_TABLE", Severity.INFO)]
+    assert all(len(line.values) == 1 for line in doc.lines)

arm_statements/tests/test_segments.py

@@ -187,3 +187,17 @@ def test_continuation_sheet_with_duplicate_date_maps_by_order(rules: Rules) -> N
     cand = excel(_npz_sheets(duplicate))
     layout = find_layout(cand, parse_header(hdr(report_date="30.09.2025"), rules), rules)
     assert layout.segments[1].value_cols == ((8, 16), (9, 21), (10, 24))
+
+
+def test_numbers_right_of_table_in_all_segments_give_one_warning(rules: Rules) -> None:
+    head: list[str | D | None] = ["Наименование", "Код", "На 30.09.2025", "На 31.12.2024", None]
+    rows: list[list[str | D | None]] = [
+        head,
+        ["Запасы", "1210", D("5"), D("4"), D("9")],
+        head,
+        ["Итого", "1200", D("5"), D("4"), D("8")],
+    ]
+    layout = find_layout(excel([("Лист1", rows)]), parse_header(hdr(), rules), rules)
+    assert [w.code for w in layout.warnings] == ["NUMBERS_OUTSIDE_TABLE"]
+    assert layout.warnings[0].locator == "Лист1!E2, Лист1!E4"
+    assert layout.segments[1].value_cols == ((2, 2), (3, 3))