607 files
157 Python scripts, 183 tests, 73 documentation files and 11 active workflow files in the supplied checkout.
Audit · 23 October 2026
SCC Air Quality: project evolution audit
Repository, data architecture, evidence pack, historical programme and publication-readiness review.
Executive assessment
Keep the evidence core. Simplify the machinery.
The project has moved beyond prototype status: its provenance, release gating, claim boundaries and public/private separation are credible. The main risk is now architectural sprawl rather than lack of capability.
2016–2025
Build one auditable ten-year measurement library before expanding the range.
Consolidate
One canonical collect → normalise → validate → analyse → report → publish path.
Protect
Scientific boundary
Keep the explicit separation between observed monitoring, modelled/satellite context, regulatory interpretation, emissions inventories and health literature.
Fix now
Reproducibility friction
Core tests should collect on the declared Python runtime without optional Google Drive dependencies being imported eagerly. Backup services must not become scientific runtime dependencies.
Simplify
Legacy AQ26 surface
Retain internal compatibility only where necessary; move public and conceptual architecture fully to SCC Air Quality and shrink the number of authoritative entry points.
Separate
Transport context
DfT annual AADF should not block or clutter the primary air-quality release. Keep transport as an optional research module tied to a defined question.
Evidence pack
2025 compliance plus measurement and research context
The supplied Defra 2025 material gives a strong statutory baseline. All zones met PM10 daily/annual limits and the PM2.5 annual limit; NO2 annual mean limits were exceeded in three zones. Defra also warns that a new national model means some differences from 2024 may reflect methodology rather than real environmental change.
NPL particle-network evidence supports a distinct particle-number, black-carbon and composition research stream. Brake/tyre and waste-incineration literature belongs in governed contextual evidence; it must not be converted by automation into source-attribution or health-harm conclusions.
Next release gates
What “externally scrutinizable” should mean.
- A clean checkout installs the declared runtime and runs core tests without secrets.
- Every public headline number is traceable to a versioned data artefact and transformation.
- Historical raw payloads are immutable and checksummed.
- Station identities and validity dates are explicit.
- Method changes are versioned separately from environmental change.
- A reviewer can reproduce one weekly release and one historical month from documented commands.