mirror of
https://github.com/Hestia-Homes/Model.git
synced 2026-06-08 11:17:27 +00:00
Contains projects associated to the development of modeling products
76e24bbdc3
Replaces the U > 3.0 W/m²K heuristic with a 3-rule cascade
discriminator that uses the BP's lodged §8 roof type alongside the
glazing type. Closes cert 000565 windows misrouting where the
previous heuristic mis-classified 3 of 6 windows.
RdSAP 10 §3.7.1 (PDF p.21) verbatim:
"Window data
Window area is assessed by measuring all windows and roof windows
throughout the dwelling. ...
Additional information to be noted: ...
• window or roof window;
• orientation"
RdSAP 10 §8.2 (PDF p.50) verbatim (Glazed walls + glazed roof):
"Glazed walls are taken as windows, glazed roof as rooflight, see
window U-values in Table 24"
The source RdSAP data set carries the "Window (vertical) / Roof
window (inclined)" classification as a discrete assessor lodgement.
The Elmhurst Summary PDF §11.0 flattens that signal — every row's
Location column reads "External wall" regardless of physical
position. The mapper must therefore reconstruct the classification.
New heuristic, in priority order:
1. "Single glazing" → never a rooflight. Approved Document L
(2006+) disallows single-glazed rooflights on energy-efficiency
grounds; SAP convention assumes Table 6c double-glazing minimum
for any (27a) entry.
2. BP roof type ∈ {"A Another dwelling above", "NR Non-residential
space above"} → rooflight. These BPs have their own structural
external roof distinct from a pitched dwelling roof — the
worksheet (30) External roof + (27a) Roof Windows treatment
follows this routing.
3. U > 3.0 W/m²K → rooflight (cohort backstop, catches cohort cert
000516 W6 Wood-frame Double pre-2002 U=3.10 on Main PA, the
only U > 3 vertical-glazing reading the cohort lodges that the
worksheet routes via (27a)).
4. Otherwise vertical.
Cohort verification: all 6 cohort certs have BPs with only PA/PN
pitched roof types (no NR/A). Rule 2 doesn't fire on cohort certs;
rule 1 doesn't block any cohort rooflights (all cohort high-U
windows are Double glazed). Rule 3 catches cohort 000516 W6
unchanged. No cohort regressions on cert→inputs cascade pins.
Cert 000565 routing fix (Summary §11.0 6-window list):
- Items 1, 6 (Main, Double, U=2.0) — vertical (unchanged)
- Item 3 (Ext1, Double, U=1.74) — vertical (unchanged; Ext1 roof
"S Same dwelling above" doesn't fire rule 2)
- Item 4 (Main, Single, U=3.35) — vertical (rule 1; was wrongly
classified as rooflight by U > 3 backstop)
- Item 2 (Ext2 NR, Triple, U=2.0) — rooflight (rule 2)
- Item 5 (Ext4 A, Double, U=2.0) — rooflight (rule 2)
Movement at HEAD `8effa2d0` → post-slice (cert 000565):
Fabric (cascade vs ws):
walls 601.22 → 602.53 (Δ -2.85 → -1.54 W/K; closes 46%)
windows 9.60 → 11.48 (Δ -1.87 → 0.00 W/K; ✓ EXACT vs ws)
roof_windows 5.02 → 3.15 (Δ +1.44 → -0.43 W/K; cascade U
formula gap exposed, see TODO below)
net fabric HTC Δ -0.99 → +0.33 W/K (magnitude improved 67%)
End-result pins:
sap_score_continuous 28.5269 → 28.4959 (Δ +0.0182 → -0.0128;
magnitude improved 30%)
ecf 5.3850 → 5.3881 (Δ -0.0016 → +0.0015)
total_fuel_cost_gbp 4678.64 → 4681.39 (Δ -1.62 → +1.13)
co2_kg_per_yr 6445.51 → 6449.13 (Δ -2.12 → +1.51)
space_heating_kwh 58980.82 → 59028.80 (Δ -27.5 → +20.5)
main_heating_fuel 34694.60 → 34722.83 (Δ -16.2 → +12.0)
lighting_kwh 1387.02 → 1382.67 (Δ +2.19 → -2.17, sign
flips: cascade DF now uses
correct rooflight area;
remaining gap is the
rooflight g×FF default-vs-
lodged drift, separate
slice)
pumps_fans_kwh ✓ EXACT (unchanged)
**Transient sap_score (integer) regression**: continuous SAP crossed
the 28.5 rounding boundary downward (28.5269 → 28.4959), so the
integer rounds to 28 instead of 29. This is a rounding artifact —
the continuous metric IS closer to ws (Δ magnitude 0.0182 → 0.0128).
Per user direction (NEXT_AGENT_PROMPT): primary metric is continuous,
transient drift OK while closing a true intermediate-value bug.
The integer pin returns to 29 once continuous SAP closes above the
ws value 28.5087.
S0380.103 cost test reframed: previously asserted total_fuel_cost
delta < +£0.05 over ws — a snapshot threshold that the SH-cascade
sign flip naturally breaks. The MEV cost split rate (12.4467
p/kWh kWh-weighted blend) is what S0380.103 specifically closes;
the test now pins that rate directly via `inputs.pumps_fans_
fuel_cost_gbp_per_kwh`, decoupled from downstream SH cascade
effects.
3-layer fix:
1. Mapper `_is_elmhurst_roof_window` predicate now takes the survey
for BP roof type lookup; new `_elmhurst_bp_roof_type` helper.
2. Two call sites at lines 327, 331 pass `survey` through.
3. New AAA test `test_summary_000565_window_routing_uses_bp_roof_
type_per_rdsap_10_section_3_7_1` pins the 4-vertical + 2-roof
classification.
Test count: 605 pass + 7 expected 000565 fails → **606 pass + 8
000565 fails** (new window-routing test + S0380.103 test reframe
both GREEN; sap_score added to work queue as a rounding-boundary
artifact). Pyright net-zero per touched file (45 baseline →
45 post-change).
Open work (in decreasing leverage on continuous SAP):
- Roof BP[1] Ext1 RR area formula refinement (+1.59 W/K over,
deferred to a separate slice per the original handover)
- Walls -1.54 W/K residual (Detailed-RR per-element investigation)
- Roof window U formula gap (-0.43 W/K; cascade formula 1/(1/U +
0.04) gives 1.852 for U_raw=2.0 but ws shows 2.1062)
- Lighting rooflight g×FF default-vs-lodged drift (-2.17 kWh)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
|
||
|---|---|---|
| .devcontainer | ||
| .github/workflows | ||
| .idea | ||
| .vscode | ||
| applications | ||
| asset_list | ||
| backend | ||
| backlog | ||
| datatypes | ||
| deployment/terraform | ||
| docs/adr | ||
| domain | ||
| epr_data_exports | ||
| etl | ||
| infrastructure | ||
| model_data/requirements | ||
| orchestration | ||
| recommendations | ||
| repositories | ||
| scripts | ||
| sfr/principal_pitch | ||
| survey_report | ||
| tests | ||
| utilities | ||
| utils | ||
| .coveragerc | ||
| .dockerignore | ||
| .gitignore | ||
| __init__.py | ||
| ara_backend_design.md | ||
| BaseUtility.py | ||
| CLAUDE.md | ||
| conftest.py | ||
| CONTEXT.md | ||
| devcontainer.sh | ||
| Dockerfile.test | ||
| Dockerfile.test.dockerignore | ||
| Makefile | ||
| MEMORY.md | ||
| package-lock.json | ||
| package.json | ||
| pyproject.toml | ||
| pyrightconfig.json | ||
| pytest.ini | ||
| README.md | ||
| run_lambda_local.sh | ||
| serverless.yml | ||
| test.requirements.txt | ||
| tox.ini | ||
| UBIQUITOUS_LANGUAGE.md | ||
Model Repository
This repository contains the code pertaining to the development of the data science and machine learning products being utilised by Hestia.
The different folders in this repository relate to services that can be used independently, or can be imported and used as part of a larger application
Getting Started
Prerequisites
Dev Container Setup
This repo uses a Docker Compose-based dev container. The model-backend service joins a shared-dev Docker network so it can communicate with other local services (e.g. a frontend container) running on your machine.
VS Code users: The initializeCommand in devcontainer.json creates the shared-dev network automatically before the container starts. No manual step required — just open the repo and select Reopen in Container.
Non-VS Code / CI workflows: Run the following once before starting the container:
make dev-setup
This is idempotent and safe to re-run if the network already exists.
Folders
backend/
This folder contains the code for the fastapi backend service, which provides an interface to much of the functionality in this repository, for the frontend
model_data/
This folder contains related to the reading and preparation of assessment model data, including pulling out epc attributes
Testing
All tests can be run, against the configuration in pytest.ini running
pytest
This will run the complete panel of tests and report on coverage in the locations specified by the pytest.ini file.
To run tests in a specific service, e.g. inside of model_data, simply run
pytest --cov-config=model_data/.coveragerc --cov=model_data
This will produce the test results and coverage reports