chore(corpus): demand-driven analyzer + confirm no generic fabric bug remains

Added scripts/corpus_1000/demand_cohort.py (isolates fabric/demand-driven
divergences from cost noise). Investigation conclusion: the demand-side tail is
one-offs, not a cluster. 12109191 (+13.1) is a LODGED decimal error (365.67 m²
floor, 10x its ground floor). 217060420 (-4.4) is spec-faithful (36.6 m2 lodged
glazing at spec U). The only >=3 structural clusters among diverging certs are
electric flats (already diagnosed spec-faithful). No generic fabric bug remains;
gauge 77.7% is near its structural ceiling. Marked 12109191 unbuildable in ledger.

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

scripts/corpus_1000/demand_cohort.py

@@ -0,0 +1,58 @@
+"""Demand-driven divergence analyzer. Flags corpus certs where the SAP miss is
+driven by FABRIC/DEMAND (PE & CO2 move with SAP), not cost — the fixable class.
+Tags nbp (building parts) and room-in-roof so multi-part / RR bugs surface."""
+from __future__ import annotations
+import json
+from pathlib import Path
+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_demand_inputs, cert_to_inputs)
+
+CORPUS = Path("backend/epc_api/json_samples/RdSAP-Schema-21.0.1/corpus.jsonl")
+
+def has_rr(doc):
+    for bp in doc.get("sap_building_parts") or []:
+        r = bp.get("room_in_roof") or bp.get("rooms_in_roof")
+        if r: return True
+    for r in doc.get("roofs") or []:
+        if "room in roof" in (r.get("description") or "").lower(): return True
+    return False
+
+rows = []
+for line in CORPUS.read_text().splitlines():
+    doc = json.loads(line)
+    L = doc.get("energy_rating_current"); peL = doc.get("energy_consumption_current")
+    co2L = doc.get("co2_emissions_current")
+    if L is None: continue
+    try:
+        epc = EpcPropertyDataMapper.from_api_response(doc)
+        r = calculate_sap_from_inputs(cert_to_inputs(epc, prices=SAP_10_2_SPEC_PRICES))
+        d = calculate_sap_from_inputs(cert_to_demand_inputs(epc, prices=SAP_10_2_SPEC_PRICES))
+    except Exception:
+        continue
+    dsap = r.sap_score_continuous - L
+    if abs(dsap) < 0.5: continue
+    peE = getattr(d, "primary_energy_kwh_per_m2", None)
+    dpe = (peE - peL) if (peE is not None and peL is not None) else None
+    co2E = d.co2_kg_per_yr/1000 if d.co2_kg_per_yr is not None else None
+    dco2 = (co2E - co2L) if (co2E is not None and co2L is not None) else None
+    nbp = len(doc.get("sap_building_parts") or [])
+    rr = has_rr(doc)
+    # demand-driven: PE moves same direction as SAP-miss inverse (PE up -> SAP down)
+    # i.e. dpe and dsap have OPPOSITE signs, and |dpe| is material
+    demand = (dpe is not None and abs(dpe) >= 4 and (dpe > 0) == (dsap < 0))
+    rows.append((doc.get("uprn"), L, round(r.sap_score_continuous,1), round(dsap,2),
+                 round(dpe,1) if dpe is not None else None,
+                 round(dco2,3) if dco2 is not None else None, nbp, rr, demand))
+
+demand_rows = [r for r in rows if r[8]]
+demand_rows.sort(key=lambda x:-abs(x[3]))
+print(f"diverging certs: {len(rows)} | demand-driven: {len(demand_rows)}")
+mp = [r for r in demand_rows if r[6] > 1]
+rr = [r for r in demand_rows if r[7]]
+print(f"  demand+multipart(nbp>1): {len(mp)} | demand+RR: {len(rr)}")
+print(f"\n=== top demand-driven divergences ===")
+print(f"{'uprn':13s} {'L':>3} {'E':>5} {'dSAP':>6} {'dPE':>6} {'dCO2':>6} {'nbp':>3} {'RR':>3}")
+for r in demand_rows[:32]:
+    print(f"{str(r[0]):13s} {r[1]:3d} {r[2]:5.1f} {r[3]:+6.2f} {str(r[4]):>6} {str(r[5]):>6} {r[6]:3d} {'Y' if r[7] else '':>3}")

scripts/corpus_1000/worklist.md

@@ -16,7 +16,7 @@ Statuses: `[ ]` todo · `[x]` resolved ≤0.5 · 🔧 fix landed · ⚠ xfail en
 - ⚠ `10012334488` · L 72 / E 85.20 / Δ +13.20 <- SPEC-FAITHFUL: SAP code 195 = electric water-storage boiler (Table 4a p.170); §12 Rule 2 bills space+water at 7-hr off-peak (5.5p). Pure cost gap (PE/CO2 match). Lodged over-billed a storage system at peak rate. Pinned engine 85 in test_real_cert_sap_accuracy.
 
 ### C004 · Σ|Δ| 13.1 · 1 certs · `hpcdb10326/gas | solid brick | pitched 75 | ext1,sec:room heaters`
-- [ ] `12109191` · L 58 / E 71.10 / Δ +13.10
+- ⛔ `12109191` · L 58 / E 71.10 / Δ +13.10 <- LODGED DATA ERROR: BP0 floor-1 area 365.67 m² (10× its 36.77 ground floor; TFA 407 for a 4-habitable mid-terrace) — a decimal misplacement. Engine faithfully computes the near-wall-less huge floor → low HTC/m² → over-rate. Garbage-in, not an engine bug.
 
 ### C005 · Σ|Δ| 10.3 · 4 certs · `h691/elec | cavity wall | (another dwelling | whw909`
 - ⚠ `line279` · L 81 / E 78.44 / Δ -2.56 <- same as 10093237426