slice 20a: ventilation_heat_loss_w_per_k feature (v2.7.0)

Adds SAP10.2 §C tracer-bullet infiltration model as a new physics-as-feature
column alongside envelope_heat_loss_w_per_k. ACH = structural baseline
(0.35 masonry / 0.25 timber-or-system-built) + open chimneys at 40 m³/h each
minus a draught-proofing reduction scaled by window_pct_draught_proofed,
then volumed and converted to W/K. Targets the d0 catastrophic-low-SAP tail
where chimney + leakage signals dominate but envelope conduction alone
under-counts heat loss.

Scope deferred to follow-ups: MVHR/MEV factors (mechanical_ventilation is
100% null in the corpus), pressure-test override (pressure_test also 100%
null - slice 18e mapper fix), open flues / passive vents / flueless gas
fires (sap_ventilation sparsely populated).

packages/domain/src/domain/ml/tests/test_transform.py

@@ -36,7 +36,7 @@ def test_transform_advertises_version_and_target_columns() -> None:
 
     # Assert
     assert isinstance(schema, TransformSchema)
-    assert schema.transform_version == "2.6.0"
+    assert schema.transform_version == "2.7.0"
     assert schema.transform_version == EpcMlTransform.VERSION
     assert set(schema.target_columns.keys()) == set(_EXPECTED_TARGET_DTYPES.keys())
     for target_name, expected_dtype in _EXPECTED_TARGET_DTYPES.items():

packages/domain/src/domain/ml/tests/test_ventilation.py

@@ -0,0 +1,149 @@
+"""Tests for ventilation_heat_loss_w_per_k.
+
+SAP10.2 §C + RdSAP10 §5 / Table 5: air change rate per hour (ACH) from
+structural infiltration + openings minus draught-proofing reduction, then
+W/K = ACH * volume_m3 * 0.33. Volume is total floor area * average storey
+height. Open chimneys, blocked chimneys, and flueless gas fires contribute
+fixed m³/h additions; draught-proofed windows reduce the structural baseline.
+"""
+
+import pytest
+
+from domain.ml.ventilation import ventilation_heat_loss_w_per_k
+
+
+def test_ventilation_bare_masonry_no_openings_returns_structural_baseline_only() -> None:
+    # Arrange — 80 m² × 2.5 m = 200 m³, masonry struct ACH = 0.35,
+    # 100% draught-proofed windows -> 0.05 reduction => 0.30 ACH total.
+    # 0.30 * 200 * 0.33 = 19.8 W/K.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=0,
+        window_pct_draught_proofed=100.0,
+    )
+
+    # Assert
+    assert result == pytest.approx(19.8, abs=0.5)
+
+
+def test_ventilation_timber_frame_no_openings_lower_struct_baseline() -> None:
+    # Arrange — same volume but timber frame -> struct ACH = 0.25.
+    # 100% DP -> 0.20 ACH; 0.20 * 200 * 0.33 = 13.2 W/K.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=True,
+        open_chimneys_count=0,
+        window_pct_draught_proofed=100.0,
+    )
+
+    # Assert
+    assert result == pytest.approx(13.2, abs=0.5)
+
+
+def test_ventilation_open_chimney_adds_40_m3_per_h() -> None:
+    # Arrange — masonry, 100% DP (0.30 base ACH), one open chimney adds 40 m³/h.
+    # In 200 m³ volume that's 0.20 ACH, so total = 0.50.
+    # 0.50 * 200 * 0.33 = 33.0 W/K.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=1,
+        window_pct_draught_proofed=100.0,
+    )
+
+    # Assert
+    assert result == pytest.approx(33.0, abs=0.5)
+
+
+def test_ventilation_no_draught_proofing_raises_structural_share() -> None:
+    # Arrange — masonry, 0% DP -> 0.35 ACH unreduced; no openings.
+    # 0.35 * 200 * 0.33 = 23.1 W/K.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=0,
+        window_pct_draught_proofed=0.0,
+    )
+
+    # Assert
+    assert result == pytest.approx(23.1, abs=0.5)
+
+
+def test_ventilation_heritage_home_with_two_chimneys_and_no_dp() -> None:
+    # Arrange — catastrophic heritage profile: TFA 100, room_h 2.5 (vol 250),
+    # masonry, 2 open chimneys (80 m³/h), 0% draught-proofed.
+    # struct=0.35, openings=80/250=0.32, total=0.67 ACH.
+    # 0.67 * 250 * 0.33 = 55.3 W/K.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=100.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=2,
+        window_pct_draught_proofed=0.0,
+    )
+
+    # Assert
+    assert result == pytest.approx(55.3, abs=1.0)
+
+
+def test_ventilation_returns_zero_when_floor_area_is_zero() -> None:
+    # Arrange — no volume, no air loss.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=0.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=1,
+        window_pct_draught_proofed=0.0,
+    )
+
+    # Assert
+    assert result == 0.0
+
+
+def test_ventilation_handles_null_chimney_count_as_zero() -> None:
+    # Arrange — open_chimneys_count is None on ~70% of the corpus; treat as 0.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=None,
+        window_pct_draught_proofed=100.0,
+    )
+
+    # Assert — same as zero-chimney case (19.8 W/K).
+    assert result == pytest.approx(19.8, abs=0.5)
+
+
+def test_ventilation_handles_null_dp_share_as_zero() -> None:
+    # Arrange — null draught-proofing share treated as no draught-proofing.
+
+    # Act
+    result = ventilation_heat_loss_w_per_k(
+        total_floor_area_m2=80.0,
+        avg_room_height_m=2.5,
+        is_timber_frame=False,
+        open_chimneys_count=0,
+        window_pct_draught_proofed=None,
+    )
+
+    # Assert — falls back to full 0.35 struct ACH -> 23.1 W/K.
+    assert result == pytest.approx(23.1, abs=0.5)

packages/domain/src/domain/ml/transform.py

@@ -35,6 +35,7 @@ from domain.ml.ecf import (
     predicted_total_fuel_cost_gbp,
 )
 from domain.ml.envelope import envelope_heat_loss_w_per_k
+from domain.ml.ventilation import ventilation_heat_loss_w_per_k
 from domain.ml.sap_efficiencies import seasonal_efficiency, water_heating_efficiency
 from domain.ml.schema import ColumnSpec, TransformSchema
 from domain.ml.ucl import apply_ucl_correction
@@ -775,6 +776,16 @@ _FEATURE_COLUMNS: dict[str, ColumnSpec] = {
             "conduction loss in W/K."
         ),
     ),
+    "ventilation_heat_loss_w_per_k": ColumnSpec(
+        dtype=float, nullable=False,
+        description=(
+            "SAP10.2 §C ventilation heat-loss in W/K from structural infiltration "
+            "(0.35 ACH masonry / 0.25 ACH timber) plus open chimneys (40 m³/h each) "
+            "minus draught-proofing reduction (0.05 max × window DP share), all "
+            "multiplied by dwelling volume × 0.33. Captures the infiltration share "
+            "of total heat loss that envelope_heat_loss_w_per_k misses. ADR-0008."
+        ),
+    ),
     "seasonal_efficiency_main_heating": ColumnSpec(
         dtype=float, nullable=False,
         description=(
@@ -902,7 +913,7 @@ class EpcMlTransform:
     Version 0.1.0 — schema contract only; feature columns added in subsequent slices.
     """
 
-    VERSION: str = "2.6.0"
+    VERSION: str = "2.7.0"
 
     def schema(self) -> TransformSchema:
         """The cross-repo ML data contract.
@@ -960,6 +971,17 @@ class EpcMlTransform:
             roof_description=_joined_descriptions(epc.roofs),
             wall_description=_joined_descriptions(epc.walls),
         )
+        main_wall_con = building_part_aggregates.get("main_dwelling_wall_construction")
+        is_timber_frame = isinstance(main_wall_con, int) and main_wall_con in (5, 6)
+        avg_room_h = building_part_aggregates.get("avg_room_height_m")
+        window_dp_pct = window_aggregates.get("window_pct_draught_proofed")
+        ventilation_w_per_k = ventilation_heat_loss_w_per_k(
+            total_floor_area_m2=epc.total_floor_area_m2,
+            avg_room_height_m=float(avg_room_h) if isinstance(avg_room_h, (int, float)) else 2.5,
+            is_timber_frame=is_timber_frame,
+            open_chimneys_count=epc.open_chimneys_count,
+            window_pct_draught_proofed=float(window_dp_pct) if isinstance(window_dp_pct, (int, float)) else None,
+        )
         main_heating_code = heating_aggregates.get("primary_sap_main_heating_code")
         water_code = heating_aggregates.get("water_heating_code")
         main_category = heating_aggregates.get("primary_main_heating_category")
@@ -1051,6 +1073,7 @@ class EpcMlTransform:
             **building_part_aggregates,
             # Features — engineered physics (ADR-0008)
             "envelope_heat_loss_w_per_k": envelope_w_per_k,
+            "ventilation_heat_loss_w_per_k": ventilation_w_per_k,
             "seasonal_efficiency_main_heating": space_eff,
             "seasonal_efficiency_water_heating": water_eff,
             "predicted_space_heating_kwh": pred_space_kwh,

packages/domain/src/domain/ml/ventilation.py

@@ -0,0 +1,56 @@
+"""Ventilation heat-loss W/K from SAP10.2 §C + RdSAP10 §5 / Table 5.
+
+The ventilation feature complements `envelope_heat_loss_w_per_k` (conduction +
+thermal bridging only). Catastrophic-low-SAP homes are dominated by
+infiltration that the conduction model can't see: open chimneys, no
+draught-proofing, leaky old windows. Tracer-bullet scope:
+
+  ACH_total = ACH_struct + (chimney_m3_h / volume_m3) − DP_reduction
+  W/K       = ACH_total × volume_m3 × 0.33
+
+Where 0.33 = ρ_air × c_p_air in kWh/(m³·K) at typical UK indoor conditions.
+
+Scope explicitly deferred:
+- Mechanical ventilation (MVHR / MEV) — `mechanical_ventilation` is 100% null
+  in the 250k corpus, so no signal to act on yet.
+- Pressure-test override — `pressure_test` is also 100% null (see slice 18e
+  candidate in HANDOFF §7-D).
+- Open flues / passive vents / extract fans / flueless gas fires — read off
+  `sap_ventilation` which itself is sparsely populated.
+"""
+
+from __future__ import annotations
+
+from typing import Final, Optional
+
+
+_STRUCTURAL_ACH_MASONRY: Final[float] = 0.35
+_STRUCTURAL_ACH_TIMBER: Final[float] = 0.25
+_DRAUGHT_PROOFING_MAX_REDUCTION: Final[float] = 0.05
+_OPEN_CHIMNEY_M3_PER_H: Final[float] = 40.0
+_AIR_HEAT_CAPACITY_KWH_PER_M3_K: Final[float] = 0.33
+
+
+def ventilation_heat_loss_w_per_k(
+    total_floor_area_m2: float,
+    avg_room_height_m: float,
+    *,
+    is_timber_frame: bool,
+    open_chimneys_count: Optional[int],
+    window_pct_draught_proofed: Optional[float],
+) -> float:
+    """SAP10.2 §C ventilation heat-loss in W/K, never null.
+
+    Linear sum: structural infiltration ACH + chimneys ACH − draught-proofing
+    reduction, multiplied by dwelling volume × 0.33.
+    """
+    if total_floor_area_m2 <= 0 or avg_room_height_m <= 0:
+        return 0.0
+    volume_m3 = total_floor_area_m2 * avg_room_height_m
+    struct_ach = _STRUCTURAL_ACH_TIMBER if is_timber_frame else _STRUCTURAL_ACH_MASONRY
+    chimney_m3_h = (open_chimneys_count or 0) * _OPEN_CHIMNEY_M3_PER_H
+    chimney_ach = chimney_m3_h / volume_m3
+    dp_share = (window_pct_draught_proofed or 0.0) / 100.0
+    dp_reduction = _DRAUGHT_PROOFING_MAX_REDUCTION * dp_share
+    total_ach = max(0.0, struct_ach - dp_reduction) + chimney_ach
+    return total_ach * volume_m3 * _AIR_HEAT_CAPACITY_KWH_PER_M3_K