fix(wall-U): as-built "insulated (assumed)" cavity uses Cavity-as-built row, not Filled cavity

An as-built cavity wall (wall_insulation_type=4) lodged "Cavity wall, as
built, insulated (assumed)" was routed to RdSAP 10 Table 6's "Filled
cavity" row. Per Table 6 (England, PDF p.41) the Filled-cavity row carries
the "†" footnote ("assumed as built") only at age bands I-M, where it is
numerically identical to "Cavity as built"; at bands A-H the Filled-cavity
row represents a GENUINE fill, not the as-built assumption. So an as-built
cavity must use the "Cavity as built" row at all bands (band G/H = 0.60,
not the filled 0.35).

This is the same latent A-H bug slice S0380.210 fixed for the "partial
insulation (assumed)" variant but left in place for "insulated (assumed)"
by a legacy production convention. The API SAP-accuracy cohort over-rated
"Cavity wall, as built, insulated (assumed)" certs at bands G/H by a clean
+1.38 / +1.61 SAP median (n=37 / n=18); bands I-M were unaffected (rows
coincide), confirming the spec mechanism per-band.

Retires the `_cavity_described_as_filled` description sniffer — as-built
cavities now always use the as-built row regardless of the rendered
insulation adjective; a genuine retrofit fill is still caught by the
explicit wall_insulation_type=2 branch.

API SAP eval: 48.6% -> 52.1% within 0.5; <1.0 63.8% -> 67.2%; median |err|
0.548 -> 0.475; mean|err| 1.561 -> 1.497; 909 computed, 0 raises.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

domain/sap10_ml/rdsap_uvalues.py

@@ -66,39 +66,26 @@ def _described_as_insulated(description: Optional[str]) -> bool:
     return "insulated" in desc or "partial insulation" in desc
 
 
-def _cavity_described_as_filled(description: Optional[str]) -> bool:
-    """True when an as-built cavity wall's description asserts the cavity is
-    insulated/filled, routing it to the Table 6 "Filled cavity" row.
-
-    Distinguishes the three as-built cavity states the EPC renders by age
-    band when wall_insulation_type=4 ("as-built / assumed"):
-
-      - "...insulated (assumed)"          → Filled cavity (assessor judges
-                                            the cavity filled but lodges no
-                                            thickness)
-      - "...partial insulation (assumed)" → "Cavity as built" row (the
-                                            as-built partial fill of the age
-                                            band, NOT a retrofit cavity fill)
-      - "...no insulation (assumed)"      → "Cavity as built" row
-
-    Narrower than `_described_as_insulated`: it excludes the "partial
-    insulation" substring so a "partial insulation (assumed)" cavity stays on
-    the as-built row. RdSAP 10 Table 6 (England) "Cavity as built" band F =
-    1.0 vs "Filled cavity" band F = 0.40 — for an as-built band-F cavity the
-    filled row understates heat loss by 2.5x. A genuine retrofit fill is
-    lodged distinctly as "Cavity wall, filled cavity"
-    (wall_insulation_type=2), handled by the explicit-code branch.
-
-    Real-cert evidence: golden cert 0390-2954-3640 (detached, band F, cavity
-    type 4, "partial insulation (assumed)") closes all four SAP metrics on
-    the as-built 1.0 row; the filled 0.40 row under-counts PE by ~28 kWh/m².
-    """
-    if description is None:
-        return False
-    desc = description.lower()
-    if "no insulation" in desc:
-        return False
-    return "insulated" in desc
+# An AS-BUILT cavity wall (wall_insulation_type=4 / "as-built / assumed",
+# however the EPC renders the insulation adjective — "insulated", "partial
+# insulation" or "no insulation" "(assumed)") routes to Table 6's "Cavity
+# as built" row via the bucketed cascade, NOT the "Filled cavity" row. Per
+# RdSAP 10 Table 6 (England) the "Filled cavity" row's † footnote ("assumed
+# as built") applies only at age bands I-M, where the two rows are
+# numerically identical — so at bands A-H the Filled cavity row represents a
+# GENUINE fill, not the as-built assumption. A genuine retrofit fill is
+# lodged distinctly as "Cavity wall, filled cavity" (wall_insulation_type=2),
+# caught by the explicit-code branch in `u_wall`.
+#
+# Slice S0380.210 first corrected this for "partial insulation (assumed)"
+# (golden 0390-2954-3640, band F → as-built 1.0); the "insulated (assumed)"
+# variant was left on the filled row by a legacy production convention. That
+# was the SAME latent A-H bug: the API SAP-accuracy cohort over-rated
+# "Cavity wall, as built, insulated (assumed)" certs at bands G/H by a clean
+# +1.4 / +1.6 SAP median (filled 0.35 vs as-built 0.60), while bands I-M
+# were unaffected (rows coincide). The `_cavity_described_as_filled`
+# description sniffer is therefore retired — as-built cavities always use the
+# as-built row regardless of the rendered insulation adjective.
 
 
 # ---------------------------------------------------------------------------
@@ -689,7 +676,6 @@ def u_wall(
         )
     if wall_type == WALL_CAVITY and (
         wall_insulation_type == WALL_INSULATION_FILLED_CAVITY
-        or _cavity_described_as_filled(description)
     ):
         return _CAVITY_FILLED_ENG[age_idx]
     bucket = _insulation_bucket(insulation_thickness_mm, insulation_present)

domain/sap10_ml/tests/test_rdsap_uvalues.py

@@ -114,20 +114,26 @@ def test_u_wall_solid_brick_with_ni_thickness_uses_50mm_row_per_table6_footnote(
     assert result == pytest.approx(0.55, abs=0.001)
 
 
-def test_u_wall_cavity_as_built_insulated_assumed_routes_to_filled_cavity_row() -> None:
-    # Arrange — 1 171 corpus certs (~4% of scanned bulk) lodge
-    # wall_insulation_type=4 ("as-built / assumed") together with the
-    # description "Cavity wall, as built, insulated (assumed)". The
-    # assessor is saying: this cavity is filled, but I haven't measured
-    # the thickness. Spec footnote on Table 6 covers this: "If a wall
-    # is known to have additional insulation but the insulation thickness
-    # is unknown, use the row in the table for 50 mm insulation" — but
-    # legacy convention (used by the production recommendation engine)
-    # is to route this to the Filled-cavity row, U = 0.7 at A-E. We
-    # follow the legacy convention here for parity with the cert assessor.
+def test_u_wall_cavity_as_built_insulated_assumed_routes_to_as_built_row() -> None:
+    # Arrange — a cavity lodged "Cavity wall, as built, insulated (assumed)"
+    # with wall_insulation_type=4 is in its AS-BUILT state, NOT a retrofit
+    # cavity fill. Per RdSAP 10 Table 6 (England) the "Filled cavity" row's
+    # † footnote ("assumed as built") applies only at bands I-M, where it
+    # coincides with "Cavity as built"; at bands A-H the filled row is for a
+    # GENUINE fill. So an as-built cavity uses the "Cavity as built" row:
+    # band E = 1.5, NOT the filled 0.7.
+    #
+    # Slice S0380.210 corrected this for the "partial insulation (assumed)"
+    # variant but left "insulated (assumed)" on the filled row by a legacy
+    # production convention — the SAME latent A-H bug. The API SAP-accuracy
+    # cohort over-rated band-G/H "insulated (assumed)" cavities by a clean
+    # +1.4 / +1.6 SAP median (filled 0.35 vs as-built 0.60); bands I-M were
+    # unaffected (rows coincide). A genuine fill lodges the distinct "Cavity
+    # wall, filled cavity" (wall_insulation_type=2), caught by the
+    # explicit-code branch.
 
     # Act
-    result = u_wall(
+    result_e = u_wall(
         country=Country.ENG,
         age_band="E",
         construction=WALL_CAVITY,
@@ -136,19 +142,30 @@ def test_u_wall_cavity_as_built_insulated_assumed_routes_to_filled_cavity_row()
         wall_insulation_type=4,
         description="Cavity wall, as built, insulated (assumed)",
     )
+    # Band I: "Cavity as built" and "Filled cavity" rows coincide (0.45),
+    # so the routing change is a no-op there — the corpus-confirmed pivot.
+    result_i = u_wall(
+        country=Country.ENG,
+        age_band="I",
+        construction=WALL_CAVITY,
+        insulation_thickness_mm=None,
+        insulation_present=False,
+        wall_insulation_type=4,
+        description="Cavity wall, as built, insulated (assumed)",
+    )
 
-    # Assert
-    assert result == pytest.approx(0.7, abs=0.001)
+    # Assert — band E → as-built 1.5 (not filled 0.7); band I → 0.45 (rows coincide).
+    assert abs(result_e - 1.5) <= 0.001
+    assert abs(result_i - 0.45) <= 0.001
 
 
 def test_u_wall_cavity_as_built_no_insulation_stays_at_table6_cavity_as_built_row() -> None:
     # Arrange — the same wall_insulation_type=4 ("as-built / assumed")
     # cert population also contains 686 "Cavity wall, as built, no
-    # insulation (assumed)" entries which must continue to route to the
-    # Cavity-as-built row of Table 6 (U=1.5 at band E). The "no
-    # insulation" substring marker takes precedence over the
-    # "insulated"-substring filled-cavity rule, so this case is
-    # disambiguated from "Cavity wall, as built, insulated (assumed)".
+    # insulation (assumed)" entries which route to the Cavity-as-built row
+    # of Table 6 (U=1.5 at band E) — as do ALL as-built cavity variants
+    # ("insulated" / "partial insulation" / "no insulation") now that the
+    # as-built path no longer special-cases the insulation adjective.
 
     # Act
     result = u_wall(
@@ -180,8 +197,8 @@ def test_u_wall_cavity_as_built_partial_insulation_routes_to_as_built_row() -> N
     # four SAP metrics on the as-built row (band F = 1.0) and under-counts
     # PE by ~28 kWh/m² on the filled row — the legacy parity was a latent
     # bug at bands A-H (bands I-M coincide per the Table 6 † footnote).
-    # The "insulated (assumed)" variant still routes to filled (see the
-    # heat_transmission `_cavity_described_as_filled` sibling test).
+    # A later slice extended the same fix to the "insulated (assumed)"
+    # variant (see the as-built-insulated sibling test above).
 
     # Act
     result = u_wall(

tests/domain/sap10_calculator/worksheet/test_heat_transmission.py

@@ -327,16 +327,22 @@ def test_solid_brick_as_built_no_insulation_assumed_stays_at_table6_as_built_row
     assert result.walls_w_per_k == pytest.approx(170.0, abs=1.0)
 
 
-def test_cavity_as_built_insulated_assumed_uses_filled_cavity_row() -> None:
-    # Arrange — the modal RdSAP encoding for a retrofitted-cavity dwelling:
-    # wall_construction=4 (cavity), wall_insulation_type=4 (as-built /
-    # assumed), and walls[0].description = "Cavity wall, as built,
-    # insulated (assumed)". The assessor has determined the cavity is
-    # filled but hasn't lodged a thickness. Without the description-based
-    # dispatcher, the cascade would return U=1.5; with it, the Filled-
-    # cavity row of Table 6 applies: U=0.7 at band E.
+def test_cavity_as_built_insulated_assumed_uses_as_built_row() -> None:
+    # Arrange — wall_construction=4 (cavity), wall_insulation_type=4
+    # (as-built / assumed), walls[0].description = "Cavity wall, as built,
+    # insulated (assumed)". This is the AS-BUILT state, not a retrofit fill:
+    # per RdSAP 10 Table 6 (England) the "Filled cavity" row's † footnote
+    # ("assumed as built") applies only at bands I-M, where it coincides
+    # with "Cavity as built"; at bands A-H the filled row is for a genuine
+    # fill. So band E uses the "Cavity as built" row U=1.5, NOT filled 0.7.
+    #
+    # Prior code special-cased the "insulated" adjective to the filled row
+    # (legacy convention); the API SAP-accuracy cohort over-rated band-G/H
+    # "insulated (assumed)" cavities by +1.4 / +1.6 SAP median (filled 0.35
+    # vs as-built 0.60). A genuine fill renders the distinct "Cavity wall,
+    # filled cavity" (wall_insulation_type=2), caught separately.
     # Geometry: 100 m² floor, 40 m perimeter, 2.5 m height, single storey
-    # → gross_wall = 100 m². walls_w_per_k expected = 0.7 × 100 = 70 W/K.
+    # → gross_wall = 100 m². walls_w_per_k expected = 1.5 × 100 = 150 W/K.
     main = make_building_part(
         identifier="Main Dwelling",
         construction_age_band="E",
@@ -367,8 +373,8 @@ def test_cavity_as_built_insulated_assumed_uses_filled_cavity_row() -> None:
     # Act
     result = heat_transmission_from_cert(epc)
 
-    # Assert
-    assert result.walls_w_per_k == pytest.approx(70.0, abs=1.0)
+    # Assert — Cavity-as-built row at band E = 1.5 W/m²K (not filled 0.7).
+    assert result.walls_w_per_k == pytest.approx(150.0, abs=1.0)
 
 
 def test_cavity_as_built_partial_insulation_assumed_uses_as_built_row() -> None:
@@ -381,8 +387,8 @@ def test_cavity_as_built_partial_insulation_assumed_uses_as_built_row() -> None:
     # RdSAP 10 Table 6 (England) "Cavity as built" band F = 1.0 vs
     # "Filled cavity" band F = 0.40. A genuine fill renders the distinct
     # "Cavity wall, filled cavity" description (wall_insulation_type=2),
-    # caught separately. Contrast the "insulated (assumed)" variant above,
-    # which the assessor judges as filled.
+    # caught separately. The "insulated (assumed)" variant above now routes
+    # to the same as-built row (all as-built adjectives coincide at A-H).
     #
     # Real-cert evidence: golden cert 0390-2954-3640 (detached, band F,
     # cavity type 4, "partial insulation (assumed)") closes all four SAP