fix(uvalues): bill known-insulation stone walls by the §5.6 thickness formula, uncapped (RdSAP 10 §5.6)

An uninsulated stone wall of lodged thickness, age bands A-E, is billed by
the RdSAP 10 §5.6 Table-12 formula on its measured thickness — sandstone /
limestone U = 54.876·W^-0.561, granite / whinstone U = 45.315·W^-0.513. Two
bugs suppressed it:

1. CAP: the as-built branch capped the formula result at the Table-6
   typical-thickness default (`if u0 >= 1.7: return 1.7`). But the formula
   only dips below 1.7 past ~488 mm (sandstone) / ~640 mm (granite), so the
   cap nullified §5.6 for essentially every real-thickness stone wall,
   under-counting fabric loss and over-rating. A measured 400 mm sandstone
   age-B wall is 1.90 (Elmhurst-confirmed), not 1.70.

2. NO INSULATION-STATE GATE: the branch fired for any stone wall with a
   lodged thickness, including ones whose insulation is "Unknown". RdSAP
   treats an unknown-insulation wall via the Table-6 typical-thickness
   default, NOT as bare stone of the lodged thickness — so a 50 mm granite
   wall with Unknown insulation must read 1.70 (worksheet), not the formula's
   6.09. Gated on `wall_insulation_type is not None`.

Fixes the 2 long-standing stone-U unit tests (granite 120 mm → 3.8871,
sandstone 120 mm → 3.7408 — they were correct red tests, not "known fails").
Corpus within-0.5 70.3% -> 71.6% (MAE 0.833 -> 0.822); ratcheted floors to
0.71 / 0.83. Worksheet fixture 000565 (granite 50 mm Unknown → 1.70) still
passes via the insulation gate. The two stone groups (sandstone/limestone vs
granite/whinstone) keep their distinct §5.6 coefficients.

pyright not installed in this codespace (strict gate not run locally).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

domain/sap10_ml/rdsap_uvalues.py

@@ -567,23 +567,25 @@ def u_wall(
     ctry = country if country is not None else Country.ENG
     age_idx = _age_index(age_band)
     band = _AGE_BANDS[age_idx]
-    # RdSAP 10 §5.6 (PDF p.40) — uninsulated stone wall thin-wall
+    # RdSAP 10 §5.6 (PDF p.40) — uninsulated stone wall formula, age
-    # formula, age bands A-E. Fires only when a documentary wall
+    # bands A-E. Fires only when (a) a documentary wall thickness is lodged
-    # thickness is lodged (per §5.3 documentary-evidence rule).
+    # (per §5.3 documentary-evidence rule) AND (b) the insulation STATE is
-    # §5.8 + Table 14 dry-line adjustment applies on top.
+    # known (`wall_insulation_type` not None — As Built / external / internal).
-    #
+    # When either is absent the cascade falls through to the Table-6
-    # Table 6 footnote (a) (PDF p.34): "Or from equations in 5.6 if
+    # typical-thickness default (1.7) below: an "insulation Unknown" lodgement
-    # the calculated U-value is less than 1.7." The cap applies only
+    # is NOT treated as bare stone of the lodged thickness (cert 000565 Ext1:
-    # to the AS-BUILT (no insulation, no dry-line) Table 6 row — for
+    # granite 50 mm + insulation Unknown → Table-6 1.70 in the worksheet, NOT
-    # thin walls where §5.6 gives U ≥ 1.7 (e.g. granite at W=50 mm
+    # the §5.6 formula's 6.09), and the footnote (a) "use §5.6 if the
-    # yields 6.09 → use Table 6 default 1.7 instead). When the wall
+    # calculated U-value is less than 1.7" clause governs the unknown path. When the thickness IS
-    # is dry-lined or insulated, the raw §5.6 result feeds the §5.8
+    # lodged the raw §5.6 U is the spec target — it is NOT capped at 1.7,
-    # chain as the input U₀ — the Table 6 footnote doesn't cap that
+    # because a thin/standard solid stone wall genuinely loses more than the
-    # path (verified empirically against cert 000565 Main alt_wall_1:
+    # typical-thickness default (sandstone 400 mm → 1.90, granite 120 mm →
-    # granite W=120 mm dry-lined → U₀=3.88 raw + dry-line → 2.34
+    # 3.89). §5.8 + Table 14 insulation / dry-line adjustments apply on top
-    # matches worksheet, NOT 1.7 + dry-line → 1.32).
+    # of the raw §5.6 U₀ (cert 000565 Main alt_wall_1: granite W=120 mm
+    # dry-lined → U₀=3.88 + dry-line → 2.34, matches worksheet).
     if (
         wall_thickness_mm is not None
+        and wall_insulation_type is not None
         and band in _STONE_AGE_A_TO_E
         and construction in (WALL_STONE_GRANITE, WALL_STONE_SANDSTONE)
     ):
@@ -628,8 +630,15 @@ def u_wall(
                         Decimal("0.01"), rounding=ROUND_HALF_UP
                     )
                 )
-            if u0 >= 1.7:
+            # As-built (uninsulated, not dry-lined) stone wall of KNOWN
-                return 1.7  # Table-6 row cap per footnote (a)
+            # thickness, age A-E: return the raw §5.6 result. The Table-6
+            # footnote (a) "< 1.7" clause governs the UNKNOWN-thickness path
+            # (which falls through to the Table-6 typical 1.7 default below) —
+            # NOT a documentary-thickness lodgement. A thin solid stone wall
+            # genuinely has U > 1.7 (e.g. sandstone 400 mm = 1.90, granite
+            # 120 mm = 3.89); capping it to 1.7 under-counts fabric loss and
+            # over-rates. Confirmed against Elmhurst (age-B sandstone 400 mm
+            # → 1.9) and the §5.6 Table-12 formula tests.
             return u0
     known_types = {
         WALL_STONE_GRANITE, WALL_STONE_SANDSTONE, WALL_SOLID_BRICK, WALL_CAVITY,

tests/infrastructure/epc_client/test_sap_accuracy_corpus.py

@@ -150,14 +150,27 @@ _CORPUS = Path(
 # MAE 0.12 -> 0.08 t/yr (bias +0.04 -> 0.00). A prior session deferred enum 9
 # ("direction not understood") while the PE/CO2 lens was confounded by the
 # climate-cascade bug (fc7c4d2d); the corrected lens shows the over-rate.
+# UNINSULATED STONE WALL §5.6 FORMULA (RdSAP 10 §5.6 Table 12, PDF p.40): a
+# stone wall of KNOWN thickness whose insulation STATE is known (As Built /
+# external / internal) is billed by the §5.6 formula on its lodged thickness
+# (sandstone/limestone U = 54.876·W^-0.561, granite/whinstone 45.315·W^-0.513),
+# NOT capped at the Table-6 typical-thickness 1.7. The old `if u0>=1.7: 1.7`
+# cap nullified the formula for every real-thickness stone wall (it only dips
+# below 1.7 past ~488 mm sandstone / ~640 mm granite) and under-counted fabric
+# loss → over-rate. Gated on `wall_insulation_type is not None` so an
+# "insulation Unknown" wall still falls to the Table-6 default (cert 000565
+# Ext1: granite 50 mm + Unknown → worksheet 1.70, not the formula's 6.09).
+# Took within-0.5 70.3% -> 71.6% (MAE 0.833 -> 0.822); fixed the 2 stone-U
+# unit tests; worksheet-validated (Elmhurst age-B sandstone 400 mm → 1.90).
+#
 # SAP RATING FLOOR (SAP 10.2 §13 / RdSAP 10 §13): the rating is floored at 1
 # ("if the result is less than 1, the rating is 1"). `calculate_sap_from_inputs`
 # now applies that floor to the CONTINUOUS score too (was integer-only), so a
 # degenerate dwelling no longer emits a negative SAP. Removed a -12.3 outlier
 # (cert 422000111926, lodged at the floor of 1, was computing -11.3): within-0.5
 # 70.2% -> 70.3%, MAE 0.845 -> 0.833.
-_MIN_WITHIN_HALF_SAP = 0.70
+_MIN_WITHIN_HALF_SAP = 0.71
-_MAX_SAP_MAE = 0.84
+_MAX_SAP_MAE = 0.83
 _MAX_CO2_MAE_TONNES = 0.09      # t CO2 / yr vs co2_emissions_current
 _MAX_PE_PER_M2_MAE = 4.0        # kWh / m2 / yr vs energy_consumption_current