Ratchet the API-corpus SAP MAE ceiling to 0.626 for the water-only C6 standing charge 🟪

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

tests/infrastructure/epc_client/test_sap_accuracy_corpus.py

@@ -367,7 +367,26 @@ _MIN_WITHIN_HALF_SAP = 0.788
 # 100031768368 SAP 59.12 -> 61.21 (lodged 61, now within 0.5); 217091901
 # 60.82 -> 61.59 (lodged 62, now an exact match). 14 corpus certs lodge
 # exactly 280mm solid brick. Unit-pinned in test_rdsap_uvalues.
-_MAX_SAP_MAE = 0.628
+# MAE 0.627 -> 0.625 (within-0.5 held at 78.8%) via the SAP 10.2 §C6
+# water-only heat-network standing charge: "Include one-half of the normal
+# heat network standing charge in the calculation of fuel costs unless the
+# space heating is also a heat network" (p.54). The half was gated on the
+# SPACE main being a network, which WHC 950/951/952 never are by
+# definition, so it could not fire. C6 phrases the half as "INCLUDE [...]
+# IN the calculation of fuel costs" against the full case's "the TOTAL
+# standing charge IS", so it is ADDITIVE to the dwelling's own Table 12
+# note (a) fuel standing rather than replacing it.
+# Blast radius is 1 cert: of the three WHC-950 certs, only 22032926 (22
+# Gwydyr Mansions, electric storage main 401) has a non-network space
+# main. £24 Economy-7 -> £24 + £60 = £84; SAP 63.737 -> 60.328 (lodged 61),
+# |err| 2.737 -> 0.672. The other two (mains 304 / 301) are on network
+# space mains and correctly keep the full £120 via the first branch.
+# NOTE: this lands on a spec argument, not an oracle — there is no
+# accredited Elmhurst worksheet for a water-only network yet (#1592-C).
+# The -0.672 overshoot is expected to move again when the C4 plant-
+# efficiency and C3.2 pumping siblings land. Unit-pinned in
+# test_cert_to_inputs.
+_MAX_SAP_MAE = 0.626
 _MAX_CO2_MAE_TONNES = 0.072     # t CO2 / yr vs co2_emissions_current
 _MAX_PE_PER_M2_MAE = 3.0        # kWh / m2 / yr vs energy_consumption_current