# Method QA Report

Generated: 2026-06-10T12:04:46+00:00

## Overall Status
- Registry, matching, summary, and model artifacts all rebuild cleanly.
- Validation passes end to end, including the mandatory Accent Drive and Te Irirangi checks.
- Key remaining caveats are local reversal dates, treatment fidelity, and uneven road-specific traffic exposure.

## Bottom-Up Checks
- Parsed source rows: 4989
- Official AT GIS reduction supplements: 1651
- Final registry rows: 4986
- Treatment-crash match rows: 36914
- Duplicate bounded source rows removed during deduping: 3
- Duplicate bounded intervention groups still remaining: 0
- Direct-source date rows with clean 24 pre / 12 post monthly panels: 3280 of 3404
- Rows using official spatial geometry: 2829

## Coverage
- descriptive_only: 2659
- high_confidence_model: 178
- manual_review: 2149

## ADT Quality
- ADT quality counts: {'none': 2313, 'high': 1651, 'low': 758, 'medium': 264}
- ADT usable-for-summary counts: {'false': 3071, 'true': 1915}
- Pre ADT age days: {'p50': 737, 'p90': 4303, 'max': 6181}
- Post ADT age days: {'p50': 379, 'p90': 1098, 'max': 2065}

## Known Roads
- Accent Drive bounded high-confidence match: True
- Te Irirangi bounded match present: True
- Te Irirangi exact high-confidence model row present: True

## Manual Review Backlog
- missing crash match: 2149

## Residual Risks
- Road-specific ADT remains uneven. Advanced road-month models use stable road baselines trended by the Auckland TMS index, with direct-ADT roads retained as a smaller sensitivity sample.
- Child injury counts remain unavailable in the current Auckland-wide crash extract and are explicitly blank in outputs.
- The supplied reduction workbook is supplemented with official AT GIS road/speed/date combinations absent from the workbook.
- Local-street reversals use the July 2025 Stage 2 completion deadline because street-level sign installation dates are not yet available.
- CAS posted-speed coding often disagrees with the programme-defined speed after changes; this may reflect partial implementation, coding lag, or remaining segment/date uncertainty.
- CAS crash-weather fields are crash-conditional and are not used as independent weather-exposure controls.
- A true Empirical Bayes analysis still requires an untreated segment inventory, segment lengths, and a calibrated Auckland safety-performance function.
- Manual review remains a supported backlog, dominated by roads with no crash-name match even before spatial filtering can be attempted.