Fix framing: lodged rating is not the accuracy target for SAP-Schema-15.0

These are 2011 SAP-2009 lodgements; this calculator implements current
RdSAP10, which has since revised several U-value tables. Closeness to
lodged was never evidence of mapper correctness (nor is a gap evidence
against it) -- the prior commit's comments and worklist entries wrongly
framed 16/20 "matching lodged" as validation and the other 4 as
anomalies needing explanation. Corrected the comments to state what the
batch actually shows: all 20 structurally distinct certs map without
error and produce plausible scores. Real accuracy validation still
needs an Elmhurst RdSAP10 rebuild on the same methodology as the
engine, which remains open.

.claude/skills/expand-sap-accuracy-corpus/worklist.md

@@ -387,30 +387,39 @@ Skip the 🚩 MVHR / 🚩 heat-pump-fuel and ⛔ sparse certs.
 - [ ] 10013924858 — SAP-17.0 · eng 81 / lodged 84
 
 # SAP-Schema-15.0 mapper validation (task a40e71c4-fd56-4a33-8677-e8643e6a5bb4)
-🔧 New schema — `from_sap_schema_15_0` added in mapper.py (PR #1531). No full
-Elmhurst rebuild done for this batch: 16/20 land exact or within 1-3 of lodged
-(strong signal alone, given each other case in this file needed a full rebuild
-to reach that confidence); the other 4 share a diagnosed, non-mapper cause
-(solid-brick U-value confirmed correct against RdSAP10 Table 6 directly — see
-comment in test_real_cert_sap_accuracy.py) so pinned without a rebuild too. All
-20 pinned as `RealCertExpectation`s.
-- [x] 🔧 100010359769 — SAP-15.0 · eng 68 (lodged 68)
-- [x] 🔧 100010359788 — SAP-15.0 · eng 72 (lodged 73)
-- [x] 🔍 100010086084 — SAP-15.0 · eng 53 (lodged 45) · solid-brick vintage-methodology gap, not mapper — see test file comment
-- [x] 🔍 100010090369 — SAP-15.0 · eng 53 (lodged 47) · ditto
-- [x] 🔍 100010074470 — SAP-15.0 · eng 46 (lodged 39) · ditto
-- [x] 🔍 100010100813 — SAP-15.0 · eng 59 (lodged 47) · ditto (largest gap, +12)
-- [x] 🔧 77179816 — SAP-15.0 · eng 76 (lodged 75)
-- [x] 🔧 10070400619 — SAP-15.0 · eng 77 (lodged 77)
-- [x] 🔧 77179820 — SAP-15.0 · eng 77 (lodged 77)
-- [x] 🔧 10070394806 — SAP-15.0 · eng 78 (lodged 77)
-- [x] 🔧 77181233 — SAP-15.0 · eng 68 (lodged 69)
-- [x] 🔧 77180626 — SAP-15.0 · eng 68 (lodged 68)
-- [x] 🔧 10070400618 — SAP-15.0 · eng 70 (lodged 69)
-- [x] 🔧 100010350459 — SAP-15.0 · eng 65 (lodged 65)
-- [x] 🔧 100010350468 — SAP-15.0 · eng 66 (lodged 66)
-- [x] 🔧 100010333531 — SAP-15.0 · eng 73 (lodged 74)
-- [x] 🔧 100010341489 — SAP-15.0 · eng 65 (lodged 64)
-- [x] 🔧 100010341491 — SAP-15.0 · eng 62 (lodged 59)
-- [x] 🔧 100010347646 — SAP-15.0 · eng 75 (lodged 75)
-- [x] 🔧 100010347637 — SAP-15.0 · eng 72 (lodged 72)
+🔧 New schema — `from_sap_schema_15_0` added in mapper.py (PR #1531). No
+Elmhurst rebuild done for this batch — and unlike every other entry in this
+file, **lodged is NOT the comparison target here**: these are 2011 SAP-2009
+lodgements, but this calculator implements current RdSAP10, which has since
+revised several U-value tables. A gap from lodged is expected either way and
+is not, by itself, evidence of a mapper bug — nor is closeness to lodged
+evidence of correctness. What this batch actually validates: all 20
+structurally distinct real certs (different fuels, meters, secondary-heating
+types, wall constructions) map without error and produce plausible scores
+through `from_sap_schema_15_0`'s three schema-wide defaults. The 4 solid-brick
+cases below show the largest gap from lodged; `u_wall` was spot-checked
+directly against RdSAP10 Table 6 for that construction/age band and returns
+the spec-correct value, so at minimum the wall U-value lookup isn't the
+source. Real accuracy validation (engine vs Elmhurst on the SAME RdSAP10
+methodology) is still open — see comment in test_real_cert_sap_accuracy.py.
+All 20 pinned as `RealCertExpectation`s to the engine's OWN observed output.
+- [x] 🔧 100010359769 — SAP-15.0 · eng 68 (2011 lodged 68, not a target)
+- [x] 🔧 100010359788 — SAP-15.0 · eng 72 (2011 lodged 73, not a target)
+- [x] 🔧 100010086084 — SAP-15.0 · eng 53 (2011 lodged 45, not a target) · solid-brick, largest-gap group; u_wall spot-checked correct
+- [x] 🔧 100010090369 — SAP-15.0 · eng 53 (2011 lodged 47, not a target) · ditto
+- [x] 🔧 100010074470 — SAP-15.0 · eng 46 (2011 lodged 39, not a target) · ditto
+- [x] 🔧 100010100813 — SAP-15.0 · eng 59 (2011 lodged 47, not a target) · ditto
+- [x] 🔧 77179816 — SAP-15.0 · eng 76 (2011 lodged 75, not a target)
+- [x] 🔧 10070400619 — SAP-15.0 · eng 77 (2011 lodged 77, not a target)
+- [x] 🔧 77179820 — SAP-15.0 · eng 77 (2011 lodged 77, not a target)
+- [x] 🔧 10070394806 — SAP-15.0 · eng 78 (2011 lodged 77, not a target)
+- [x] 🔧 77181233 — SAP-15.0 · eng 68 (2011 lodged 69, not a target)
+- [x] 🔧 77180626 — SAP-15.0 · eng 68 (2011 lodged 68, not a target)
+- [x] 🔧 10070400618 — SAP-15.0 · eng 70 (2011 lodged 69, not a target)
+- [x] 🔧 100010350459 — SAP-15.0 · eng 65 (2011 lodged 65, not a target)
+- [x] 🔧 100010350468 — SAP-15.0 · eng 66 (2011 lodged 66, not a target)
+- [x] 🔧 100010333531 — SAP-15.0 · eng 73 (2011 lodged 74, not a target)
+- [x] 🔧 100010341489 — SAP-15.0 · eng 65 (2011 lodged 64, not a target)
+- [x] 🔧 100010341491 — SAP-15.0 · eng 62 (2011 lodged 59, not a target)
+- [x] 🔧 100010347646 — SAP-15.0 · eng 75 (2011 lodged 75, not a target)
+- [x] 🔧 100010347637 — SAP-15.0 · eng 72 (2011 lodged 72, not a target)

tests/domain/sap10_calculator/test_real_cert_sap_accuracy.py

@@ -1091,24 +1091,36 @@ _EXPECTATIONS: Final[tuple[RealCertExpectation, ...]] = (
     # SAP-Schema-15.0 campaign (task a40e71c4-fd56-4a33-8677-e8643e6a5bb4,
     # portfolio 824 / scenario 1278 — the failing task that surfaced this
     # schema gap; see `from_sap_schema_15_0` in mapper.py). 20 real 2011-era
-    # LIG-15.0 lodgements sampled across the failing portfolio. 16 of 20 land
-    # exact or within 1-3 of the lodged score, confirming the mapper's three
-    # schema-wide defaults (door_count=1, WWHRS/FGHRS "none fitted",
-    # percent_draughtproofed=0) are sound.
+    # LIG-15.0 lodgements sampled across the failing portfolio.
     #
-    # The remaining 4 (086084/090369/074470/100813) share solid-brick
-    # UNINSULATED walls, age band C, and diverge from lodged by +6 to +12 —
-    # every other affected field (fuel, meter type, secondary heating type)
-    # varies independently across them, so wall construction is the only
-    # correlate. Checked `u_wall(Country.ENG, "C", 3, None,
-    # wall_insulation_type=4)` directly: it returns 1.70, RdSAP10's own
-    # published Table 6 value for solid 220-280mm brick in that age band — no
-    # mapper/U-value bug found. These are 2011 lodgements, assessed under the
-    # older SAP-2009 methodology's (higher/more conservative) solid-wall
-    # U-values; re-scoring under current RdSAP10 legitimately lands higher.
-    # Same category as the documented full-SAP-through-RdSAP and reduced-
-    # field party-wall vintage gaps elsewhere in this file — PINNED to the
-    # engine's observed value, not lodged; not tuned.
+    # NOTE on what these pins do and don't validate: `sap_score` here is the
+    # ENGINE'S OWN observed output, not an accuracy claim against the lodged
+    # `energy_rating_current`. These are 2011 certs, assessed under the SAP
+    # 2009 methodology of the day; this calculator implements the CURRENT
+    # RdSAP10 methodology, which has since revised several U-value tables
+    # (solid-wall thickness bands among them — see `u_wall` §5.7 Table 13).
+    # A 2011-vs-2026-methodology delta is therefore EXPECTED and tells you
+    # nothing about mapper correctness either way — closeness to lodged is
+    # not evidence for the mapper, and a gap from lodged is not evidence
+    # against it. Real accuracy validation needs an Elmhurst RdSAP10 rebuild
+    # (the `validate-cert-sap-accuracy` skill) on the SAME methodology as the
+    # engine, which this batch did not do.
+    #
+    # What this batch DOES validate: all 20 structurally distinct real certs
+    # (different fuels, meter types, secondary-heating types, wall
+    # constructions, portfolios) map without error through the new
+    # `from_sap_schema_15_0` path and produce plausible SAP scores — i.e. the
+    # three schema-wide defaults (door_count=1, WWHRS/FGHRS "none fitted",
+    # percent_draughtproofed=0) don't crash or produce nonsense on a diverse
+    # sample. Four of the twenty (086084/090369/074470/100813, sharing
+    # solid-brick UNINSULATED walls / age band C) show the largest gap from
+    # lodged; spot-checked `u_wall(Country.ENG, "C", 3, None,
+    # wall_insulation_type=4)` directly and it returns 1.70, matching RdSAP10's
+    # own published Table 6 for that construction/age band — so at minimum
+    # the wall U-value lookup itself is not the source of any divergence.
+    # PINNED to the engine's observed value on every case (the file's
+    # standing convention) — not tuned to lodged, and not claimed as an
+    # accuracy match either.
     RealCertExpectation(
         schema="SAP-Schema-15.0",
         sample="uprn_100010359769",