Slice 47: Summary_000474 chain pins SAP at 1e-4 vs worksheet PDF

Two diffs closed against the hand-built `_elmhurst_worksheet_000474`
target (SAP 62.2584):

1. `pumps_fans_kwh_per_yr` (130 → 160). The cascade keys §4f pumps+fans
   electricity on `MainHeatingDetail.main_heating_category` (gas-fired
   boilers = cat 2 → 160 kWh/yr). `from_elmhurst_site_notes` wasn't
   populating the field, so it fell through to the default 130. Added
   `_elmhurst_main_heating_category` deriving cat 2 for the gas/LPG-
   PCDB-boiler branch; other categories deferred until a fixture
   exercises them (consistent with the cascade lookup).

2. Window [4] orientation `East-South` → `East` and window [5]
   orientation `''` → `South-East`. The layout-style parser's
   `before_start = prev_manuf + 7` / `after_end = next_data` rule was
   over-grabbing prefix tokens of W_{k+1} as suffix tokens of W_k
   ('South' from W_5's prefix bled into W_4's suffix). Replaced with
   a symmetric partition on the first glazing-type-start token
   (`Single`/`Double`/`Triple`/`Secondary`) within the cross-window
   gap, used as the upper bound of W_k's suffix and the lower bound
   of W_{k+1}'s prefix. Same boundary on both sides — prefix tokens
   of the next window can no longer be attributed as suffix of the
   current one.

After both fixes, Summary_000474 → ElmhurstSiteNotes → EpcPropertyData
→ cascade → SAP matches the worksheet PDF's unrounded line 257 value
to 1e-4 tolerance. All 754 datatypes/epc/ + backend/documents_parser/
tests green; pyright net-zero on touched files.

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

backend/documents_parser/elmhurst_extractor.py

@@ -424,6 +424,15 @@ class ElmhurstSiteNotesExtractor:
         "North", "South", "East", "West", "NE", "NW", "SE", "SW",
     })
     _BP_INLINE_TOKENS = frozenset({"Main"})  # "Extension" only appears as suffix
+    # The Elmhurst Summary PDF lodges each window's glazing-type as a
+    # capitalised phrase like "Double between 2002" / "Double with unknown"
+    # / "Single" / "Triple" / "Secondary". The first token of that phrase
+    # marks the start of a new window's prefix block in the layout dump,
+    # which is the only stable signal partitioning one window's suffix
+    # from the next window's prefix.
+    _GLAZING_TYPE_PREFIX_WORDS = frozenset({
+        "Single", "Double", "Triple", "Secondary",
+    })
 
     def _extract_windows_from_layout(self) -> List[Window]:
         """Fallback window parser for Summary PDFs preprocessed from
@@ -457,21 +466,34 @@ class ElmhurstSiteNotesExtractor:
             manuf_idx = self._find_manufacturer_after(lines, data_idx)
             if manuf_idx is None:
                 continue
-            prev_window_end = (
+            prev_manuf_idx = (
-                self._estimate_window_end(lines, data_anchors[k - 1][0])
+                self._find_manufacturer_after(lines, data_anchors[k - 1][0])
-                if k > 0 else 0
+                if k > 0 else None
             )
-            next_window_start = (
+            next_data_idx = (
                 data_anchors[k + 1][0] if k + 1 < len(data_anchors) else len(lines)
             )
+            # Partition the cross-window gap between this window's suffix
+            # and the next window's prefix on the first glazing-type-start
+            # token (Single/Double/Triple/Secondary). The same boundary
+            # is used symmetrically — current window's `after_end` = next
+            # window's `before_start` — so prefix tokens of W_{k+1} never
+            # get attributed as suffix of W_k (which was the bug producing
+            # orientation='East-South' for windows where 'South' actually
+            # belonged to the next row).
+            before_start = (
+                self._partition_after_manuf(lines, prev_manuf_idx, data_idx)
+                if prev_manuf_idx is not None else 0
+            )
+            after_end = self._partition_after_manuf(lines, manuf_idx, next_data_idx)
             try:
                 window = self._parse_window_from_anchors(
                     lines=lines,
                     data_idx=data_idx,
                     manuf_idx=manuf_idx,
                     anchor=anchor,
-                    before_start=prev_window_end,
+                    before_start=before_start,
-                    after_end=next_window_start,
+                    after_end=after_end,
                 )
             except (ValueError, IndexError):
                 continue
@@ -485,15 +507,22 @@ class ElmhurstSiteNotesExtractor:
                 return j
         return None
 
-    def _estimate_window_end(self, lines: List[str], data_idx: int) -> int:
+    def _partition_after_manuf(
-        """End-of-window index (exclusive) for the window whose data
+        self, lines: List[str], manuf_idx: int, next_data_idx: int
-        line is at `data_idx`. Used to bound the "before" segment of
+    ) -> int:
-        the *next* window when extracting suffix tokens."""
+        """Return the exclusive upper bound for this window's suffix
-        manuf_idx = self._find_manufacturer_after(lines, data_idx)
+        block (and the inclusive lower bound for the next window's prefix
-        if manuf_idx is None:
+        block). After the manufacturer line come 3 fixed tokens (g_value,
-            return data_idx + 1
+        draught, shutters); the variable suffix lines start at manuf+4
-        # Manufacturer + g_value + draught + shutters + ~3 suffix tokens
+        and run until the next window's glazing-type-start token (e.g.
-        return manuf_idx + 7
+        'Double between 2002', 'Single', 'Triple ...') or until the
+        next window's data line if no such token is present."""
+        scan_start = manuf_idx + 4
+        for j in range(scan_start, next_data_idx):
+            first_word = lines[j].strip().split(" ", 1)[0]
+            if first_word in self._GLAZING_TYPE_PREFIX_WORDS:
+                return j
+        return next_data_idx
 
     def _parse_window_from_anchors(
         self,

datatypes/epc/domain/mapper.py

@@ -1,6 +1,6 @@
 import re
 from datetime import date
-from typing import List, Optional, Sequence, Union, Dict, Any
+from typing import Any, Dict, Final, List, Optional, Sequence, Union
 from datatypes.epc.schema.helpers import from_dict
 
 from datatypes.epc.domain.epc_property_data import (
@@ -62,6 +62,7 @@ from datatypes.epc.surveys.elmhurst_site_notes import (
     BuildingPartDimensions as ElmhurstBuildingPartDimensions,
     ElmhurstSiteNotes,
     FloorDetails as ElmhurstFloorDetails,
+    MainHeating as ElmhurstMainHeating,
     RoofDetails as ElmhurstRoofDetails,
     VentilationAndCooling as ElmhurstVentilation,
     WallDetails as ElmhurstWallDetails,
@@ -2220,6 +2221,34 @@ def _elmhurst_sap_control_code(sap_control: str) -> Optional[int]:
     return int(m.group(1)) if m else None
 
 
+# SAP10.2 Table 4a main-heating-category codes. Currently only the
+# gas-fired-boiler branch is exercised by the Elmhurst cohort — the
+# cascade reads `main_heating_category` to key the §4f pumps+fans table
+# (160 kWh/yr for cat 2 = 115 central heating pump + 45 flue fan) and to
+# detect heat-network mains (cat 6). Other categories (heat pumps,
+# warm-air, electric storage, oil/biomass) are deferred until a fixture
+# exercises them.
+_ELMHURST_HEATING_CATEGORY_GAS_BOILER: Final[int] = 2
+_ELMHURST_GAS_BOILER_FUEL_TYPES: frozenset[str] = frozenset({
+    "Mains gas",
+    "LPG bottled",
+    "LPG bulk",
+    "LPG special condition",
+})
+
+
+def _elmhurst_main_heating_category(
+    mh: ElmhurstMainHeating, pcdb_index: Optional[int]
+) -> Optional[int]:
+    """Derive the SAP10.2 Table 4a main-heating-category from Elmhurst-
+    lodged data. A PCDB-referenced boiler on mains/LPG gas is category 2
+    (gas-fired boilers); other system types fall through to None so the
+    cascade applies its default pumps_fans 130 kWh/yr until extended."""
+    if pcdb_index is not None and mh.fuel_type in _ELMHURST_GAS_BOILER_FUEL_TYPES:
+        return _ELMHURST_HEATING_CATEGORY_GAS_BOILER
+    return None
+
+
 def _map_elmhurst_sap_heating(survey: ElmhurstSiteNotes) -> SapHeating:
     mh = survey.main_heating
     sap_control = mh.heating_controls_sap
@@ -2241,6 +2270,7 @@ def _map_elmhurst_sap_heating(survey: ElmhurstSiteNotes) -> SapHeating:
     main_fuel_int = _elmhurst_main_fuel_int(mh.fuel_type)
     heat_emitter_int = _elmhurst_heat_emitter_int(mh.heat_emitter)
     sap_control_int = _elmhurst_sap_control_code(sap_control)
+    main_heating_category = _elmhurst_main_heating_category(mh, pcdb_index)
     return SapHeating(
         instantaneous_wwhrs=InstantaneousWwhrs(),
         main_heating_details=[
@@ -2256,6 +2286,7 @@ def _map_elmhurst_sap_heating(survey: ElmhurstSiteNotes) -> SapHeating:
                 fan_flue_present=mh.fan_assisted_flue,
                 main_heating_control=sap_control_int if sap_control_int is not None else control,
                 central_heating_pump_age_str=mh.heat_pump_age,
+                main_heating_category=main_heating_category,
                 # Per RdSAP, a PCDB-listed boiler is data source 1
                 # (manufacturer measured efficiency); the integer index
                 # number drives PCDB lookup in the cascade.