# Handover — post S0380.189 (thermal mass parameter / Table 22)

Point-in-time note. Start from [`AGENT_GUIDE.md`](AGENT_GUIDE.md) for the
methodology, accuracy bar, and pipeline — this doc only records *what this
session did* and *what is open*.

- **Branch:** `feature/per-cert-mapper-validation`
- **HEAD:** `e03f08cd` (S0380.189)
- **Baseline:** `2290 passed, 1 skipped, 0 failed` (the skip is the
  gitignored xlsx `load_cells` test). Verify with the §4 suite command.

---

## What this session shipped (S0380.185–189)

| Slice | What | Spec |
|---|---|---|
| **.185** | Recorded the CH6 "pin-forever" proof — distribution-loss is an Elmhurst Summary-export gap, not a mapper miss (controlled adjoining-dwellings pair byte-identical in inputs). | — |
| **.186** | Added `test_golden_cert_pe_co2_matches_worksheet` — pins calc PE/CO2 for the 47 worksheet-backed certs against the dr87 `(286)`/`(272)` at full precision (not lodged register values). | Appendix U |
| **.187** | Appendix M1 §3a `D_PV,m` was missing **electric secondary** space heating `(215)m` — under-credited PV on gas-main+electric-secondary+PV certs. | SAP 10.2 App M1 §3a |
| **.188** | `D_PV,m` used the Appendix L **L12 lighting GAIN** (`= E_L×0.85`) instead of the **L10 lighting ELECTRICITY** `(232)`. Closed the whole PV cohort to 1e-4. Same gain-vs-electricity class as the S0380.73 cooking fix. | SAP 10.2 App L L10/L12, M1 §3a |
| **.189** | **Thermal mass parameter** was hardcoded 250 at all 5 §7/§8 call sites. Now `_thermal_mass_parameter_kj_per_m2_k(epc)` per Table 22. | RdSAP 10 §5.16 Table 22 (p.48) |

After .186–.188 **all 47 worksheet-backed certs match calc≡worksheet at
<1e-4 on PE and CO2** (the convergence target). See
[`project-golden-coverage-state` memory] for the per-slice detail.

---

## The S0380.189 diagnosis (template for the open work)

Driven entirely by the **per-line walk** (AGENT_GUIDE §3) against a
**user-simulated worksheet** for a 6035-archetype property:

- **Fixture:** `sap worksheets/golden fixture debugging/simulated case 1/`
  — `Summary_001431 (1).pdf` (input) + `P960-0001-001431 - ….pdf` (worksheet
  ground truth). A gas-combi mid-terrace, TFA 128, solid brick **with
  internal insulation**, no PV/secondary/cylinder. (The `P960-…` prefix is
  just the Elmhurst account id; cert is 001431.) **These PDFs are untracked**
  — the user holds them locally.
- **Decompose:** PE +8.78 / SAP −1.76 / CO2 +0.21, *entirely* space-heating
  demand (+838 kWh). Fabric `(26–37)`, internal gains `(73)`, climate
  `(96)m`, HTC `(39)` all EXACT → localised to **§7 MIT** `(92)` +0.71 °C.
- **Root cause:** TMP hardcoded 250; cert is masonry **with internal
  insulation** → RdSAP 10 §5.16 Table 22 = **100**. Wrong TMP → time
  constant τ=Cm/(3.6·H) ≈40 h not 16 h → §7 temperature reduction too small
  → MIT too high → space heating over-stated.
- **Fix + blast radius:** only golden cert **6035** re-pinned (SAP −6→−2, PE
  +46.42→+19.16, CO2 +1.07→+0.42). All other fixtures are masonry-no-internal
  → TMP unchanged.

### Two diagnostic traps that cost false starts (read before debugging §7/§8)

1. **The worksheet has TWO blocks per dwelling.** The first is the SAP-rating
   block (UK-average climate, region 0); the second, under *"CALCULATION OF
   EPC COSTS, EMISSIONS AND PRIMARY ENERGY"*, is the postcode block. The
   **demand cascade (`cert_to_demand_inputs`) matches the POSTCODE block.**
   Comparing calc HTC/MIT to the UK-avg block shows phantom gaps (e.g. HTC
   "−10.8"); against the postcode block HTC is exact.
2. **`(84)` Total gains = internal `(73)` + solar `(83)`.** The calc's
   `internal_gains_annual_avg_w` is `(73)` only — don't diff it against `(84)`
   (a phantom "−248 W" gap that is just solar).

Use the §2.4 section helpers (`mean_internal_temperature_section_from_cert`,
`space_heating_section_from_cert`, `internal_gains_section_from_cert`,
`local_climate_for_cert`) for the per-line walk.

---

## OPEN — next slices (ranked)

### 1. Summary-path `main_fuel_type` derivation from the SAP code  ← do first
The Elmhurst **Summary PDF has no main-heating fuel field** — only the SAP
code (`14.0 Main Heating1 → Main Heating SAP Code 104`). So
`from_elmhurst_site_notes` leaves `main_fuel_type=''`, and `cert_to_inputs`
raises `MissingMainFuelType` (cert_to_inputs.py `_main_fuel_code`). This
**blocks the Summary path for every gas-combi cert**, including the simulated
case (I injected `main_fuel_type=26` to run the diagnosis).

Fix: derive the fuel in the mapper (or `_main_fuel_code`) from
`sap_main_heating_code` via SAP 10.2 Table 4b (104 = condensing combi **mains
gas**), mirroring the existing strict-raise → derive pattern. Cite the table.
This is the higher-leverage win — it unblocks the whole site-notes gas-combi
population, not one cert.

### 2. Pin the simulated 001431 case end-to-end (after #1)
Once #1 lands, the Summary path runs natively. Add it as a site-notes fixture
and pin every line ref at 1e-4. **NB:** with .189 + injected fuel, PE and CO2
close to 1e-4 but **SAP showed +0.0007** vs a hand-computed target (`100 −
13.95×ECF` from the 4-dp `(257)`=1.6047). That is almost certainly ECF
rounding in the target, not a real gap — but **verify against the worksheet's
own continuous SAP** before declaring it closed (don't trust my hand value).

### 3. Cert 6035 remaining +19 PE
6035 is API/lodged-only (no worksheet) so it can't be pinned past the lodged
register. The user can reproduce 6035 *exactly* in Elmhurst to get a
worksheet — offer to format the golden JSON
(`tests/domain/sap10_calculator/rdsap/fixtures/golden/6035-7729-2309-0879-2296.json`)
as Elmhurst inputs. With a worksheet, walk the remaining +19 the same way.

### Carry-over (lower priority)
- `transform.py:973` treats `wall_construction in (5,6)` as timber-frame for
  the ventilation structural-ACH split, but Table 22 / `rdsap_uvalues`
  classify 6 = **system built (masonry)**, only 5/7/8 are timber/cob/park.
  Possible latent ventilation-ACH bug — verify before touching.

---

## Process notes
- One slice = one commit, spec citation in the message, `Co-Authored-By:
  Claude Opus 4.8` trailer. AAA tests, `abs(x-y) <= tol` (not `pytest.approx`).
- The golden worksheet PE/CO2 pins (.186) re-pin smaller as gaps close — never
  widen. The lodged-residual pins (`test_golden_cert_residual_matches_pin`)
  carry the API-vs-register residual and move when the calc improves.