Slice S0380.83: Extractor + mapper recognise Exposed / Connected gable_type per RdSAP 10 §3.10

The Elmhurst Summary PDF §8.1 "Room(s) in Roof" per-surface table publishes the
gable-wall environment column with one of four values:

  Party                          → §8.1 party-wall row
  Sheltered                      → §8.1 sheltered external row
  Exposed                        → §8.1 exposed external row
  Connected (to heated space)    → §8.1 internal partition

Per RdSAP 10 §3.10 (PDF p.30-35) "Detailed Room-in-Roof" + Table 4 (p.22)
"Heat-loss surface variants":

  - Exposed gable wall → external wall at the lodged U-value
  - Sheltered gable wall → external wall at the lodged U-value
  - Party gable wall → party wall at U=0.25 (Table 4 row 2)
  - Connected gable wall → internal partition to heated space, NOT a
    heat-loss surface

The extractor was only capturing `gable_type ∈ {"Party", "Sheltered",
"Connected to heated space"}` — neither `"Exposed"` (every external gable
on cert 000565) nor the plain `"Connected"` string (the actual PDF
lodging value, vs the verbose "Connected to heated space" form used on
other Summary schemas) was recognised. Both fell through with
`gable_type=None`, masking the downstream cascade gap (cert 000565
BP[0] Main Gable Wall 1 is lodged "Exposed" at U=0.35 but extracted
as untyped → mapper routes to `gable_wall` party at U=0.25, vs the
worksheet's "Roof room Main Gable Wall 1" at U=0.35).

This slice closes the extractor side only:

  backend/documents_parser/elmhurst_extractor.py:_parse_rir_surface_row
  expands its `gable_type` lookup set to include "Exposed" and the
  plain "Connected" lodging value.

Mapper-side: `_map_elmhurst_rir_surface` (datatypes/epc/domain/mapper.py)
preserves cert 9501's behaviour — its flat-RR elif previously hinged
on `surface.gable_type is None and is_flat`; now extends to
`surface.gable_type in (None, "Exposed") and is_flat` so the same
flat-RR routing fires whichever lodging shape the Summary PDF uses.

Net cascade impact: zero. Cert 9501 (top-floor flat) retains its
RR-gables-as-external routing. Cert 000565 (house) keeps falling
through to the default `gable_wall` (party at U=0.25) routing for
"Exposed" + "Connected" gables — the next slice in the block reroutes
those to external walls + drops Connected surfaces per RdSAP 10
Table 4. This commit is pure data-extraction completion; pin
movement lands when S0380.84 wires the mapper through.

Test baseline: 555 pass + 8 expected `test_sap_result_pin[000565-*]`
fails (was 554 + 8 at S0380.82; one new test pins the spec rule).
Pyright net-zero on touched files (45 errors, matches baseline).

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

backend/documents_parser/elmhurst_extractor.py

@@ -523,7 +523,10 @@ class ElmhurstSiteNotesExtractor:
                     insulation = t
             elif t in ("Mineral or EPS", "PUR", "PIR"):
                 insulation_type = t
-            elif t in ("Party", "Sheltered", "Connected to heated space"):
+            elif t in (
+                "Party", "Sheltered", "Exposed",
+                "Connected", "Connected to heated space",
+            ):
                 gable_type = t
         return RoomInRoofSurface(
             name=name,

backend/documents_parser/tests/test_summary_pdf_mapper_chain.py

@@ -509,6 +509,79 @@ def test_summary_9501_rr_gable_walls_route_to_external_walls_hlc() -> None:
     assert abs(ht.walls_w_per_k - worksheet_walls_w_per_k) <= 1e-2
 
 
+def test_summary_000565_extractor_recognises_exposed_and_connected_gable_types() -> None:
+    """Summary PDF §8.1 Room(s) in Roof per-surface table lists the
+    gable-wall environment column with one of four published values:
+
+        Party                          → §8.1 party-wall row
+        Sheltered                      → §8.1 sheltered external row
+        Exposed                        → §8.1 exposed external row
+        Connected (to heated space)    → §8.1 internal partition
+
+    Per RdSAP 10 §3.10 (PDF p.30-35) "Detailed Room-in-Roof" + Table 4
+    (p.22) "Heat-loss surface variants":
+
+      - Exposed gable wall → external wall at the lodged U-value (or
+        the BP main-wall U when the lodged value is the default)
+      - Sheltered gable wall → external wall at the lodged U-value
+      - Party gable wall → party wall at U=0.25 (Table 4 row 2)
+      - Connected gable wall → internal partition to heated space,
+        NOT a heat-loss surface (drops from external + party totals)
+
+    The extractor was only capturing `gable_type ∈ {"Party",
+    "Sheltered", "Connected to heated space"}` — neither `"Exposed"`
+    (every external gable on cert 000565) nor the plain `"Connected"`
+    string (the actual lodging used in Summary PDFs vs the verbose
+    "Connected to heated space") was recognised. Both fell through
+    with `gable_type=None`, masking the downstream cascade gap (cert
+    000565 BP[0] Main Gable Wall 1 is lodged "Exposed" at U=0.35 but
+    extracted as untyped → mapper routes to `gable_wall` (party at
+    U=0.25) — see worksheet "Roof room Main Gable Wall 1" line at
+    U=0.35).
+
+    This pin asserts the extractor surfaces the lodged environment
+    column verbatim. The downstream mapper + cascade behaviour stays
+    unchanged until follow-up slices use the new field — this is a
+    pure extractor data-completion step (no test pins move).
+    """
+    # Arrange
+    pages = _summary_pdf_to_textract_style_pages(_SUMMARY_000565_PDF)
+    site_notes = ElmhurstSiteNotesExtractor(pages).extract()
+
+    # Act — Main BP gables; Ext1/Ext2 gables expose both "Connected"
+    # and "Exposed" values from the cert lodging.
+    rir_main = site_notes.room_in_roof
+    main_surfaces = {s.name: s for s in (rir_main.surfaces if rir_main else [])}
+    rir_ext1 = (
+        site_notes.extensions[0].room_in_roof
+        if site_notes.extensions and len(site_notes.extensions) > 0
+        else None
+    )
+    ext1_surfaces = {s.name: s for s in (rir_ext1.surfaces if rir_ext1 else [])}
+
+    # Assert
+    # Main BP[0]: Gable Wall 1 lodged "Exposed" (default U 0.35); Gable
+    # Wall 2 lodged "Sheltered" (default U 0.30).
+    assert main_surfaces["Gable Wall 1"].gable_type == "Exposed", (
+        f"Main Gable Wall 1 gable_type = "
+        f"{main_surfaces['Gable Wall 1'].gable_type!r}; expected 'Exposed'"
+    )
+    assert main_surfaces["Gable Wall 2"].gable_type == "Sheltered", (
+        f"Main Gable Wall 2 gable_type = "
+        f"{main_surfaces['Gable Wall 2'].gable_type!r}; expected 'Sheltered'"
+    )
+    # Ext1 BP[1]: Gable Wall 1 lodged "Connected" (internal partition);
+    # Gable Wall 2 lodged "Exposed" (default U 1.70).
+    assert ext1_surfaces["Gable Wall 1"].gable_type == "Connected", (
+        f"Ext1 Gable Wall 1 gable_type = "
+        f"{ext1_surfaces['Gable Wall 1'].gable_type!r}; expected 'Connected'"
+    )
+    assert ext1_surfaces["Gable Wall 2"].gable_type == "Exposed", (
+        f"Ext1 Gable Wall 2 gable_type = "
+        f"{ext1_surfaces['Gable Wall 2'].gable_type!r}; expected 'Exposed'"
+    )
+
+
 def test_summary_9501_pv_array_surfaced_from_elmhurst_section_19() -> None:
     # Arrange — cert 9501's Elmhurst §19.0 PV section lodges measured
     # array detail (2.36 kWp, South-West orientation, 45° elevation,

datatypes/epc/domain/mapper.py

@@ -3277,12 +3277,19 @@ def _map_elmhurst_rir_surface(
     if kind == "gable_wall" and surface.gable_type == "Sheltered":
         kind = "gable_wall_external"
         u_value_override = surface.default_u_value
-    elif kind == "gable_wall" and surface.gable_type is None and is_flat:
+    elif (
+        kind == "gable_wall"
+        and surface.gable_type in (None, "Exposed")
+        and is_flat
+    ):
         # Flat with RR: gables are external by default (top of block,
         # no neighbour above). Lodge as gable_wall_external with no
         # u_value override so the cascade falls through to the main-
         # wall U (`uw` in heat_transmission.py:674) — matches cert
         # 9501's worksheet treatment of both gable walls at U=1.7.
+        # Per Summary PDF schema the gable env column reads "Exposed"
+        # for the same case the legacy heuristic detected via None;
+        # both lodging shapes route here.
         kind = "gable_wall_external"
     area_m2 = _round_half_up_2dp(surface.length_m, surface.height_m)
     if kind in ("gable_wall", "gable_wall_external"):