Wire the guarded conversational RAG answer layer end-to-end

ingestion/tests/test_segment_detector.py

@@ -1,5 +1,9 @@
 from ingestion.extract.models import Span
-from ingestion.segment.detector import detect_monograph_titles, detect_section_headings
+from ingestion.segment.detector import (
+    detect_monograph_titles,
+    detect_section_headings,
+    in_monograph_range,
+)
 
 
 def _span(text, physical_page, printed_page, y0=100.0, bold=True, size=10.0):
@@ -86,3 +90,12 @@ def test_unknown_bold_text_not_matched_as_section():
 def test_section_heading_outside_monograph_range_excluded():
     spans = [_span("Chỉ định", 5, 6, bold=True, size=9.5)]
     assert list(detect_section_headings(spans)) == []
+
+
+def test_back_index_cannot_reenter_range_via_bad_inferred_printed_page():
+    # Confirmed real failure: physical page 1655 of the back index was mapped
+    # to printed page 1496, making its "Tương tác thuốc" entry extend
+    # ZOLPIDEM's source range from page 1494 through page 1655.
+    index_span = _span("Tương tác thuốc", 1655, 1496)
+    assert in_monograph_range(index_span) is False
+    assert list(detect_section_headings([index_span])) == []