nwau-py

nwau_py

This package provides utilities for working with IHACPA funding weights.

Installation

The project is standardizing on uv for local development and validation. Install and sync the environment with:

uv sync --locked --group dev --group test --group coverage --group typing --group property --group mutation --group profiling --group docs

The SAS calculators for each pricing year should be extracted under archive/sas/<YEAR>/ so the modules can read the reference tables.

Development targets Python 3.10 through 3.14. The current quality-gate stack uses uv, Ruff, pytest, pytest-cov, ty, Codecov, Hypothesis, mutmut, Scalene, and on-demand Vale alongside the calculator libraries. ty is the active type checker in this phase; mypy remains only as a transitional comparator while the migration is completed. Codecov is fed by the XML coverage artifact produced in CI so the same report can be reviewed in the UI and in local runs.

Rust contributors should also run cargo fmt --all --check, cargo clippy --all-targets --all-features -- -D warnings, and cargo test against the workspace under rust/.

The slow-validation workflow is split between a weekly scheduled run and a manual workflow_dispatch run. Its profiling job writes Scalene output under .cache/validation/scalene/, which keeps generated reports outside version control while still making them easy to inspect locally.

The calculator implementation still uses pandas in places, while the longer-term data stack is moving toward Arrow-backed interchange and Polars where parity work allows it.

Data preparation

The Python calculators load tables directly from the SAS releases archived under archive/sas/<YEAR>/. If you need the original Excel workbook outputs, the helper script excel_calculator/scripts/extract_all.py can recreate weights.csv and formula.json from the workbooks stored in excel_calculator/archive/<year>.

CLI usage

Once installed you can calculate funding from the command line using the nwau_py CLI:

python -m nwau_py.cli.main acute patient_data.csv --output funding.csv --year 2025

patient_data.csv should contain the variables expected by the SAS programs. The output CSV will include an NWAU25 column with the calculated weights.

Validation Workflow

The package is validated through manifest-driven fixture packs and explicit parity checks rather than broad claims of complete coverage. Use the shared fixture helpers in nwau_py.fixtures to load packs, read payloads, and run runner-neutral checks that can later be consumed by Python, C#, and web tooling.

For the acute 2025 pilot there is also an explicit Rust-backed opt-in path, nwau_py.calculators.calculate_acute_rust_2025(), which loads the optional nwau_py._rust extension through nwau_py.rust_bridge and returns the same NWAU25 values for the synthetic fixture pack without changing the default Python calculator path.

Formula JSON

The file excel_calculator/data/formula.json is a structured representation of the workbook formula used by IHACPA and is kept for archival purposes. It lists each symbol (for example PW, APaed, AInd) and the corresponding column heading in excel_calculator/data/weights.csv. A full mapping is provided in FORMULA_MAPPING.md.

The steps array enumerates the intermediate calculations that build up the final NWAU25 value. Starting from the base price weight, paediatric and other adjustments are applied, ICU hours are added, private patient adjustments and readmission penalties are subtracted and the result is multiplied by the National Efficient Price (NEP). The last step yields the overall NWAU25 figure used for funding.

Documentation

Detailed notes on each calculator module are available in docs/calculators.md.

Golden Fixture Packs

The project uses manifest-driven fixture packs for cross-language parity checks. The current pilot pack lives under tests/fixtures/golden/acute_2025/ and includes a JSON manifest plus input.csv and expected.csv payloads.

Use nwau_py.fixtures to work with these packs:

• load_fixture_pack() validates the manifest and resolves payload paths.
• discover_fixture_packs() finds packs under a fixture root.
• read_payload_frame() loads the manifest-declared input or expected output tables.
• iter_fixture_cases() maps manifests to calculator cases.
• fixture_case_params() turns cases into pytest parameters with stable ids.
• iter_fixture_pytest_params_from_root() discovers packs and generates pytest parameters from manifest metadata.
• run_fixture_case() executes a calculator against one manifest case.
• run_fixture_suite() executes and validates a set of cases.
• run_fixture_suite_from_root() discovers packs and runs every valid case in order.
• assert_fixture_case_output() checks the result against the declared tolerance and rounding policy.

Pytest coverage is generated from the same manifests by parameterizing over the discovered fixture cases. That keeps the test matrix and the helper runner in sync without hard-coding runner-specific logic.