Slice 51: Summary_000516 chain pins SAP at 1e-4; roof-window separation

Three mapper extensions, validated by 000516 closing to 1e-4:

1. Roof-window separation by U-value threshold. Elmhurst Summary PDFs
   pool roof windows into the §11 vertical-window table with no type
   marker. The U-value is the only reliable signal — vertical glazing
   in the cohort tops out at 2.80 W/m²K, while Table 24 roof windows
   start at 3.0+. `_is_elmhurst_roof_window` filters U > 3.0 into
   `sap_roof_windows`; the rest flow through the `sap_windows` path.

2. Table-24 roof-window U-value lookup. The cohort lodges Manufacturer
   U=3.10 for the 000516 roof window, but the worksheet's (27a) line
   (U_eff=2.99) reverse-engineers to a raw U=3.40 — the RdSAP10
   Table 24 "Double pre 2002" roof-window default. `_elmhurst_roof_
   window_u_value` keyed on glazing-type captures the +0.3 W/m²K step;
   falls back to the lodged U for glazing types not yet in the table.

3. `SapWindow.window_width × window_height = lodged Area` convention.
   The Elmhurst Summary PDF carries lodged W (2 d.p.) × lodged H
   (2 d.p.) AND a precomputed Area (2 d.p., not always equal to
   product after rounding). The cascade reads only the W×H product
   across §3 / §5 / §6, so flattening to `(area, 1.0)` keeps the
   downstream area aligned with the worksheet's rounded value rather
   than reconstructing W×H with its own rounding drift (e.g. 1.22 ×
   1.76 = 2.1472 m² vs lodged 2.15 m²). The existing
   `test_first_window_*` tests pinning literal W/H were updated to
   pin the area product (the cascade-relevant invariant).

Cohort state after this slice:

  000474   0.0000  ✓ Slice 47
  000477  +1.1161     Elmhurst floor_ach quirk
  000480   0.0000  ✓ Slice 50
  000487  +1.1844     extractor still drops most §11 windows
  000490   0.0000  ✓ Slice 49
  000516   0.0000  ✓ THIS SLICE

4/6 closed at 1e-4. 756 tests pass; pyright net-zero (35 baseline).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/documents_parser/tests/test_elmhurst_end_to_end.py

@@ -201,11 +201,20 @@ class TestWindows:
     def test_window_count(self, result: EpcPropertyData) -> None:
         assert len(result.sap_windows) == 4
 
-    def test_first_window_width(self, result: EpcPropertyData) -> None:
-        assert result.sap_windows[0].window_width == 1.30
+    def test_first_window_area(self, result: EpcPropertyData) -> None:
+        # The Elmhurst mapper lodges the Summary PDF's precomputed Area
+        # (1.30 × 1.10 = 1.43 m²) as `window_width × 1.0` to avoid the
+        # 2-d.p. round-trip drift that W × H reintroduces. The cascade
+        # reads only the product, so flattening to (area, 1.0) is
+        # behaviourally equivalent to (1.30, 1.10) modulo precision.
+        w = result.sap_windows[0]
+        assert w.window_width * w.window_height == 1.43
 
     def test_first_window_height(self, result: EpcPropertyData) -> None:
-        assert result.sap_windows[0].window_height == 1.10
+        # See `test_first_window_area` — the mapper normalises height
+        # to 1.0 so the lodged Area can be carried as the canonical
+        # geometry without re-multiplying.
+        assert result.sap_windows[0].window_height == 1.0
 
     def test_first_window_orientation(self, result: EpcPropertyData) -> None:
         # SAP10 octant code: 1 = North. The solar-gains cascade keys

backend/documents_parser/tests/test_summary_pdf_mapper_chain.py

@@ -41,6 +41,7 @@ _FIXTURES = Path(__file__).parent / "fixtures"
 _SUMMARY_000474_PDF = _FIXTURES / "Summary_000474.pdf"
 _SUMMARY_000480_PDF = _FIXTURES / "Summary_000480.pdf"
 _SUMMARY_000490_PDF = _FIXTURES / "Summary_000490.pdf"
+_SUMMARY_000516_PDF = _FIXTURES / "Summary_000516.pdf"
 
 
 def _summary_pdf_to_textract_style_pages(pdf_path: Path) -> list[str]:
@@ -160,6 +161,28 @@ def test_summary_000480_full_chain_sap_matches_worksheet_pdf_exactly() -> None:
     assert abs(result.sap_score_continuous - worksheet_unrounded_sap) < 1e-4
 
 
+def test_summary_000516_full_chain_sap_matches_worksheet_pdf_exactly() -> None:
+    # Arrange — cert U985-0001-000516 is a mid-terrace with main bp +
+    # 19.02 m² room-in-roof. Worksheet PDF lodges unrounded SAP 62.7937.
+    # The §11 table mixes 5 vertical windows (U=2.80) with 1 roof
+    # window (U=3.10 in cert, U=3.40 Table 24 raw); the mapper
+    # discriminates by `U > 3.0` and routes the high-U entry to
+    # `sap_roof_windows` so its solar gains feed §6 with the right
+    # pitch (45°) and Table-24 U-value.
+    pages = _summary_pdf_to_textract_style_pages(_SUMMARY_000516_PDF)
+    site_notes = ElmhurstSiteNotesExtractor(pages).extract()
+    epc = EpcPropertyDataMapper.from_elmhurst_site_notes(site_notes)
+
+    # Act
+    result = calculate_sap_from_inputs(
+        cert_to_inputs(epc, prices=SAP_10_2_SPEC_PRICES)
+    )
+
+    # Assert
+    worksheet_unrounded_sap = 62.7937
+    assert abs(result.sap_score_continuous - worksheet_unrounded_sap) < 1e-4
+
+
 def test_summary_000490_full_chain_sap_matches_worksheet_pdf_exactly() -> None:
     # Arrange — cert U985-0001-000490 is an end-terrace with main +
     # 1st extension. The worksheet PDF lodges unrounded SAP 57.3979.

datatypes/epc/domain/mapper.py

@@ -22,6 +22,7 @@ from datatypes.epc.domain.epc_property_data import (
     SapFlatDetails,
     SapFloorDimension,
     SapHeating,
+    SapRoofWindow,
     SapRoomInRoof,
     SapRoomInRoofSurface,
     SapVentilation,
@@ -283,7 +284,14 @@ class EpcPropertyDataMapper:
             main_heating=[],
             door_count=survey.door_count,
             sap_heating=_map_elmhurst_sap_heating(survey),
-            sap_windows=[_map_elmhurst_window(w) for w in survey.windows],
+            sap_windows=[
+                _map_elmhurst_window(w) for w in survey.windows
+                if not _is_elmhurst_roof_window(w)
+            ],
+            sap_roof_windows=[
+                _map_elmhurst_roof_window(w) for w in survey.windows
+                if _is_elmhurst_roof_window(w)
+            ] or None,
             sap_energy_source=SapEnergySource(
                 mains_gas=survey.meters.main_gas,
                 meter_type=survey.meters.electricity_meter_type,
@@ -2262,6 +2270,55 @@ def _elmhurst_orientation_int(orientation: str) -> int:
     return _ELMHURST_ORIENTATION_TO_SAP10.get(orientation, 1)
 
 
+# SAP10.2 §3.2 / Table 24: roof windows have higher U-values than
+# vertical glazing of the same age — typically U >= 3.0 W/m²K vs
+# vertical-glazing 2.0–2.8. The Elmhurst Summary PDF doesn't lodge
+# a discrete "window type" field, so we use the lodged U-value as
+# the discriminator. None of the six cohort certs has a vertical
+# window > 2.8 W/m²K; the only U=3.10 entry (000516 W5, 1.18 m²,
+# matching the U985 worksheet's "Roof Windows 1(Main)" row) is the
+# correct positive — and falling through to a vertical window
+# misallocates its solar gains + applies the wrong Table-6c U.
+_ELMHURST_ROOF_WINDOW_U_THRESHOLD: Final[float] = 3.0
+
+
+def _is_elmhurst_roof_window(w: ElmhurstWindow) -> bool:
+    """Heuristic discriminator: roof windows have U-value > 3.0 in the
+    Elmhurst cohort. The Summary PDF doesn't carry an explicit type
+    flag; the U985 worksheet PDFs separate them into a distinct
+    `Roof Windows N(Main)` row in §3, matching the U-threshold here."""
+    return w.u_value > _ELMHURST_ROOF_WINDOW_U_THRESHOLD
+
+
+# RdSAP10 Table 24 — roof-window U-value by glazing-type cell as lodged
+# in the Elmhurst Summary PDF §11 row. The lodged "Manufacturer 3.10"
+# rate is the vertical-glazing reading; Elmhurst applies the roof-window
+# Table-24 default (+0.3 W/m²K typical step) when scoring against the
+# worksheet's (27a) line. The cohort exercises only "Double pre 2002".
+_ELMHURST_ROOF_WINDOW_U_BY_GLAZING: Dict[str, float] = {
+    "Double pre 2002": 3.4,
+}
+
+
+def _elmhurst_roof_window_u_value(w: ElmhurstWindow) -> float:
+    """Roof-window U-value per RdSAP10 Table 24 — keyed on the lodged
+    glazing-type phrase. Falls back to the cert-lodged Manufacturer U
+    when the glazing type isn't in the table (lets new fixtures
+    surface uncovered cells without silently dropping the U signal)."""
+    return _ELMHURST_ROOF_WINDOW_U_BY_GLAZING.get(w.glazing_type, w.u_value)
+
+
+def _map_elmhurst_roof_window(w: ElmhurstWindow) -> SapRoofWindow:
+    return SapRoofWindow(
+        area_m2=w.area_m2,
+        u_value_raw=_elmhurst_roof_window_u_value(w),
+        orientation=_elmhurst_orientation_int(w.orientation),
+        pitch_deg=45.0,
+        g_perpendicular=w.g_value,
+        frame_factor=w.frame_factor,
+    )
+
+
 def _map_elmhurst_window(w: ElmhurstWindow) -> SapWindow:
     return SapWindow(
         frame_material=w.frame_type or None,
@@ -2269,8 +2326,15 @@ def _map_elmhurst_window(w: ElmhurstWindow) -> SapWindow:
         orientation=_elmhurst_orientation_int(w.orientation),
         window_type="Window",
         glazing_type=w.glazing_type,
-        window_width=w.width_m,
-        window_height=w.height_m,
+        # SapWindow's width × height is consumed across §3 (windows_w_per_
+        # k), §5 (daylight factor), and §6 (solar gains) — all summed as
+        # the area product. The Elmhurst Summary PDF lodges W and H to
+        # 2 d.p. then a precomputed Area to 2 d.p.; using width=area /
+        # height=1.0 keeps the cascade aligned with the worksheet's
+        # rounded area rather than reconstructing W×H with its own
+        # rounding error (e.g. 1.22 × 1.76 = 2.1472 vs lodged 2.15).
+        window_width=w.area_m2,
+        window_height=1.0,
         draught_proofed=w.draught_proofed,
         window_location=w.building_part,
         window_wall_type=w.location,