fix(rdsap): thin present wall-insulation no longer maps to uninsulated bucket

Field-level bias scan (scripts/corpus_1000/field_bias.py) flagged solid-brick +
internal-insulation certs (wall_con+ins=(3,3)) systematically under-rating
(median -0.40, n=18). Root cause: _insulation_bucket routed insulation that is
KNOWN PRESENT but with a thin lodged thickness (5-24mm, below the smallest
tabulated 50mm column) to bucket 0 (uninsulated, U 1.7) - WORSE than the
present-but-unknown case, which the RdSAP Table 6 footnote correctly routes to
the 50mm row. A known 10mm shouldn't read as higher-U than not knowing at all.

Now present insulation never falls to bucket 0; thin/unknown thickness both
round up to the 50mm row. Only genuinely absent insulation maps to 0.

Corpus cert 5117280 (10mm internal, no wall thickness -> bucket path): Δ -4.44
-> +1.08. Corpus MAE 0.641 -> 0.637, PE MAE 3.0 -> 2.9. Gauge 77.7% holds; 44
RealCert pins + 342 fabric tests green. TDD: 2 new unit tests.

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

domain/sap10_ml/rdsap_uvalues.py

@@ -353,7 +353,13 @@ def _insulation_bucket(thickness_mm: Optional[int], insulation_present: bool) ->
     to the 50 mm bucket when `insulation_present=True`; when not
     present, the as-built (bucket 0) row applies regardless.
     """
-    if insulation_present and (thickness_mm is None or thickness_mm == 0):
+    # Insulation known to be present must never fall to the uninsulated
+    # (bucket 0) row. RdSAP 10 Table 6 footnote routes present-but-unknown
+    # thickness to the 50 mm row; a KNOWN thin lodgement (5-24 mm, below the
+    # smallest tabulated 50 mm column) must not read as WORSE (higher U) than
+    # the unknown case — so it too rounds up to the 50 mm row rather than to
+    # bucket 0. Only genuinely absent insulation (present=False) maps to 0.
+    if insulation_present and (thickness_mm is None or thickness_mm < 25):
         return 50
     if thickness_mm is None:
         return 0

domain/sap10_ml/tests/test_rdsap_uvalues.py

@@ -31,6 +31,7 @@ from domain.sap10_ml.rdsap_uvalues import (
     WALL_STONE_SANDSTONE,
     WALL_SYSTEM_BUILT,
     WALL_TIMBER_FRAME,
+    _insulation_bucket,
     thermal_bridging_y,
     u_door,
     u_exposed_floor,
@@ -114,6 +115,43 @@ 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_solid_brick_present_insulation_thin_known_thickness_not_uninsulated() -> None:
+    # Arrange — a solid-brick wall lodging internal insulation (type 3)
+    # with a KNOWN thin thickness (10 mm) but no documentary wall
+    # thickness (so the §5.8 formula branch, which needs wall_thickness,
+    # cannot fire) falls to the Table-6 bucket. A known 10 mm value sits
+    # below the smallest tabulated 50 mm column, but insulation IS present
+    # — it must NOT read as WORSE (higher U) than the present-but-unknown
+    # case, which the footnote routes to the 50 mm row. Pre-fix,
+    # `_insulation_bucket(10, present=True)` returned 0 (uninsulated,
+    # U=1.7), under-crediting the wall (corpus cert 5117280: Δ −4.44).
+
+    # Act
+    result = u_wall(
+        country=Country.ENG,
+        age_band="B",
+        construction=WALL_SOLID_BRICK,
+        insulation_thickness_mm=10,
+        insulation_present=True,
+        wall_insulation_type=3,
+    )
+
+    # Assert — present insulation → 50 mm row (U=0.55), never uninsulated 1.7.
+    assert result == pytest.approx(0.55, abs=0.001)
+
+
+def test_insulation_bucket_present_thin_thickness_rounds_up_to_50_not_zero() -> None:
+    # Direct guard on the bucket: present insulation never maps to bucket 0.
+    assert _insulation_bucket(10, True) == 50
+    assert _insulation_bucket(24, True) == 50
+    assert _insulation_bucket(None, True) == 50
+    assert _insulation_bucket(50, True) == 50
+    assert _insulation_bucket(100, True) == 100
+    # Genuinely absent insulation still routes thin/zero to bucket 0.
+    assert _insulation_bucket(10, False) == 0
+    assert _insulation_bucket(None, False) == 0
+
+
 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

scripts/corpus_1000/field_bias.py

@@ -0,0 +1,46 @@
+"""Field-level systematic-bias scan: mean Δ(SAP) grouped by each mapped
+attribute. A subgroup with large |meanΔ| AND enough n is a systematic mapping
+error for that attribute (would cancel in the overall ~0 bias)."""
+from __future__ import annotations
+import json, statistics
+from collections import defaultdict
+from datatypes.epc.domain.mapper import EpcPropertyDataMapper
+from domain.sap10_calculator.calculator import calculate_sap_from_inputs
+from domain.sap10_calculator.rdsap.cert_to_inputs import SAP_10_2_SPEC_PRICES, cert_to_inputs
+CORPUS="backend/epc_api/json_samples/RdSAP-Schema-21.0.1/corpus.jsonl"
+G=defaultdict(lambda: defaultdict(list))
+def gv(x): return x.get('value') if isinstance(x,dict) else x
+for line in open(CORPUS):
+    d=json.loads(line); L=d.get('energy_rating_current')
+    if L is None: continue
+    try:
+        epc=EpcPropertyDataMapper.from_api_response(d)
+        r=calculate_sap_from_inputs(cert_to_inputs(epc,prices=SAP_10_2_SPEC_PRICES))
+    except Exception: continue
+    dsap=r.sap_score_continuous-L
+    bp=(d.get('sap_building_parts') or [{}])[0]
+    sh=d.get('sap_heating') or {}; md=(sh.get('main_heating_details') or [{}])[0]
+    G['wall_con'][bp.get('wall_construction')].append(dsap)
+    G['wall_ins'][bp.get('wall_insulation_type')].append(dsap)
+    G['wall_con+ins'][(bp.get('wall_construction'),bp.get('wall_insulation_type'))].append(dsap)
+    G['roof_con'][bp.get('roof_construction')].append(dsap)
+    G['floor_hl'][bp.get('floor_heat_loss')].append(dsap)
+    G['age'][bp.get('construction_age_band')].append(dsap)
+    G['main_cat'][md.get('main_heating_category')].append(dsap)
+    G['whc'][sh.get('water_heating_code')].append(dsap)
+    G['meter'][(d.get('sap_energy_source') or {}).get('meter_type')].append(dsap)
+    G['country'][d.get('country_code')].append(dsap)
+    G['dwelling'][d.get('dwelling_type')].append(dsap)
+    G['nbp'][len(d.get('sap_building_parts') or [])].append(dsap)
+    G['glz'][tuple(sorted(set(w.get('glazing_type') for w in (d.get('sap_windows') or []))))[:1]].append(dsap)
+print("=== subgroups with |meanΔ| >= 0.5 and n >= 8 (systematic-bias candidates) ===")
+hits=[]
+for field, groups in G.items():
+    for key, vals in groups.items():
+        if len(vals) >= 8:
+            m=statistics.mean(vals)
+            if abs(m) >= 0.5:
+                hits.append((abs(m), field, key, len(vals), m, statistics.median(vals)))
+for am,field,key,n,m,med in sorted(hits, reverse=True):
+    print(f"  {field:12s} = {str(key):20s} n={n:4d} meanΔ={m:+.2f} medianΔ={med:+.2f}")
+if not hits: print("  (none — no systematic subgroup bias)")

scripts/corpus_1000/worklist.md

@@ -73,7 +73,7 @@ Statuses: `[ ]` todo · `[x]` resolved ≤0.5 · 🔧 fix landed · ⚠ xfail en
 - [ ] `10093342182` · L 82 / E 77.51 / Δ -4.49
 
 ### C022 · Σ|Δ| 4.4 · 1 certs · `hpcdb17973/gas | solid brick | (another dwelling`
-- [ ] `5117280` · L 76 / E 71.56 / Δ -4.44
+- [ ] `5117280` · L 76 / E 71.56 / Δ -4.44  <- 🔧 thin-internal-insulation bucket-0 fix: present insulation + known <25mm thickness was mapped to uninsulated (U 1.7); now routes to 50mm row like unknown-thickness. Δ-4.44 -> +1.08.
 
 ### C023 · Σ|Δ| 4.4 · 1 certs · `h191/elec | solid brick | (another dwelling | ext2`
 - [ ] `217060420` · L 42 / E 37.58 / Δ -4.42