Record ADR-0049: ASHP sized to the dwelling's design heat loss at the efficiency peak

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

docs/adr/0049-ashp-sized-to-the-dwelling-design-heat-loss.md

@@ -0,0 +1,98 @@
+# The ASHP recommendation sizes the heat pump to the dwelling's design heat loss
+
+## Status
+
+accepted
+
+## Context
+
+The ASHP measure (ADR-0024) installed **one fixed heat pump** — a 4.37 kW Vaillant
+aroTHERM plus (PCDB Table 362 record 110257) — on every eligible dwelling, as a
+"representative, contractor-installable" end-state. SAP 10.2 Appendix N reads a
+heat pump's efficiency at the dwelling's **PSR** (Power-to-Space-heating-Ratio =
+rated output / design heat loss): efficiency peaks around PSR 0.8 and, per
+footnote 44, **collapses toward 100% below the record's smallest PSR**.
+
+A portfolio-796 audit found the fixed pump scoring the ASHP **below the baseline**
+on high-heat-loss dwellings. Property 712794 (uprn 10002468116): a leaky dwelling
+with a **17.5 kW design heat loss**, so a 4.37 kW pump lands at **PSR 0.25** — the
+efficiency collapse zone. The calculator correctly applied ~195% SPF; combined
+with pricing electricity at the standard tariff against an off-peak storage
+baseline, the ASHP scored **SAP 10.33 vs the baseline 11.96 (−1.63)** and the
+Optimiser never selected it. Every electric-heated dwelling too leaky for a 4.37 kW
+pump was capped below the goal band by this — a modelling defect, not a calculator
+one (SAP's PSR curve is correct: a grossly undersized heat pump *is* inefficient).
+
+The fix needs the dwelling's design heat loss, which the calculator computes
+internally (annual-average HLC × the SAP design temperature difference, 24.2 K)
+but did not surface.
+
+Sizing raised a second question the accredited data answered. MCS capacity sizing
+targets **PSR ≈ 1.0** (output ≥ design heat loss). But the **relodged Elmhurst ASHP
+cert** — a real dwelling our calculator gives a 6.40 kW design heat loss — was
+fitted with the **5 kW** Vaillant (110257, 4.37 kW output), i.e. **PSR 0.68**, and
+scores SAP 72.08. A PSR-1.0 rule picks the 7 kW pump for that dwelling → SAP 73.86,
+**oversizing by a rung and inflating SAP by ~1.8**, and no longer reproduces the
+accredited cert. SAP efficiency at PSR 0.68 and 1.0 is near-identical (~400%), so
+the real installer's smaller, cheaper choice is the realistic one.
+
+## Decision
+
+Size the heat pump to the dwelling and expose what sizing needs:
+
+1. **`SapResult.design_heat_loss_kw`** — the calculator surfaces the design heat
+   loss (annual-average HLC × 24.2 K / 1000), the quantity Appendix N's PSR
+   divides the pump output by.
+
+2. **A curated aroTHERM plus ladder** — one representative PCDB record per nominal
+   size across the domestic range (`_ASHP_SIZING_LADDER`): 3.5 kW (110249), 5 kW
+   (110257, the prior anchor), 7 kW (110265), 10 kW (110273), 12 kW (110281). All
+   the same "& AI VIH RW" variant, so the ladder is a consistent product line.
+
+3. **Size to the Appendix-N efficiency peak, not MCS capacity.**
+   `select_ashp_pcdb_id` picks the rung whose rated output is nearest **0.8 ×
+   design heat loss** (PSR ≈ 0.8, the efficiency peak). This reproduces the pump a
+   real installer fits — validated against the relodged Elmhurst cert at delta 0 —
+   and keeps the pump clear of the low-PSR collapse. A load beyond the 12 kW range
+   naturally selects the largest rung (an honest, capped undersize; those
+   dwellings need fabric-first).
+
+4. **Cost and efficiency share one figure.** `ashp_cost_inputs` takes the same
+   design heat loss (over the floor-area proxy), so the rate-sheet install band
+   (ADR-0025) and the pump's efficiency record are sized from one number. The
+   orchestrator reads `design_heat_loss_kw` off a baseline score and threads it
+   into the heating generator; the harness computes it via a calculator pass.
+
+## Consequences
+
+- Property 712794's ASHP goes from **−1.63 SAP (never selected)** to **+44.2 SAP
+  (selected)**; its plan lifts **G (12.0) → B (89.1)**. Electric-heated dwellings
+  previously capped below goal C by the undersized pump can now reach it — the
+  largest HIGH-severity mover on the portfolio-796 `plan-stops-short-of-goal` and
+  `plan-stuck-in-low-band` groups.
+- The accredited relodged Elmhurst ASHP cert still reproduces at **delta 0** (the
+  6.40 kW dwelling sizes to the 5 kW Vaillant it was actually fitted with). Two
+  self-snapshot cascade pins re-pin to their correctly-sized larger pumps (both
+  dwellings were undersized by the fixed unit, so their end-state SAP rises).
+- Adding `design_heat_loss_kw` to `SapResult` is a required field; manual
+  constructions (test stubs) carry it.
+- Extends [ADR-0024](0024-ashp-bundle-absolute-end-state.md) (the ASHP end-state)
+  and [ADR-0025](0025-ashp-cost-composed-from-the-rate-sheet.md) (size-banded
+  cost); the cost bands already scaled — they now receive the real heat loss.
+
+### Alternatives rejected
+
+- **MCS capacity sizing (PSR ≈ 1.0, output ≥ design heat loss).** Rejected: it
+  oversizes relative to real installs, inflating SAP ~1.8 per dwelling and
+  breaking the accredited-cert reproduction. SAP efficiency is equivalent at
+  PSR 0.68–1.0, so the realistic (installer) size is preferred.
+- **Keep the fixed 4.37 kW pump.** Rejected: the collapse on high-heat-loss
+  dwellings is the defect under repair.
+- **Fabric-first sequencing (score the ASHP only after insulation).** A real
+  concern — the Optimiser scores each measure independently against the
+  un-insulated baseline (ADR-0016) — but a separate change to measure scoring/
+  sequencing, not pump sizing. Sizing the pump fixes the standalone score
+  regardless; noted for follow-up.
+- **Nearest record from all ~10,000 Table 362 heat pumps.** Rejected: an opaque,
+  unauditable selection; a curated single-product ladder keeps ADR-0024's
+  "representative, contractor-installable" intent.