Model/packages
Khalim Conn-Kowlessar c496f345f8 slice 14l: bigger-run fixes — UCL guard, PV Measurement coercion, sMAPE
Three changes surfaced by the 25k 2026 run:
- transform._peui_ucl returns None for non-positive raw PEUI (net-exporters).
  apply_ucl_correction would otherwise raise ValueError on negative input.
- PhotovoltaicArray scalars (peak_power, pitch, orientation, overshading)
  now accept Measurement | int | float in the schema; mapper coerces via
  _measurement_value.
- train_baseline reports sMAPE alongside MAPE — handles zero-actual rows
  (e.g. co2_emissions for net-zero certs) where MAPE explodes.

Results at N=25,000 RdSAP 2026 certs (~32s end-to-end):
  sap_score          MAPE=0.064  sMAPE=0.054  R^2=0.762
  co2_emissions      sMAPE=0.140  R^2=0.890
  peui_raw           MAPE=0.126  sMAPE=0.120  R^2=0.714
  peui_ucl           MAPE=0.114  sMAPE=0.108  R^2=0.736
  space_heating_kwh  MAPE=0.167  sMAPE=0.157  R^2=0.915
  hot_water_kwh      MAPE=0.089  sMAPE=0.086  R^2=0.737

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
2026-05-16 21:15:37 +00:00
..
domain slice 14l: bigger-run fixes — UCL guard, PV Measurement coercion, sMAPE 2026-05-16 21:15:37 +00:00
fetchers added potential file scaffolding: 2026-05-15 10:56:53 +00:00
repos added potential file scaffolding: 2026-05-15 10:56:53 +00:00
utils added potential file scaffolding: 2026-05-15 10:56:53 +00:00
README.md added potential file scaffolding: 2026-05-15 10:56:53 +00:00

Shared packages

Workspace packages consumed by services/*. Each package is its own Python distribution with its own pyproject.toml; services import via the workspace dependency mechanism ({ workspace = true }).

Package Purpose
domain/ Shared domain types — Property, BaselinePerformance, Plan, Scenario, EpcPropertyData, etc. No persistence, no IO, no business logic.
repos/ Persistence layer — one repo per aggregate. Owns the SQL. Depends on domain.
fetchers/ External API clients (gov EPC, Ofgem, Google Solar, etc.). Depend on domain for response shapes.
utils/ Cross-cutting infra — logging, S3, CloudWatch URL builders, SQS task helpers.

Adding a new shared package

Only when a real second consumer materialises. Don't pre-shatter (repos-epc, repos-property, ...) — split when a deployment needs to drop a dep, not before.

See ../ara_backend_design.md §11 for the broader monorepo layout and ../CONTEXT.md for the domain glossary that names the types living in domain/.