Visual inspection of municipal road pavements under the Dutch CROW 146b guideline does not scale to the 130,013 km of municipal and water-board road network in the Netherlands (CBS, 2025). Existing AI methods report mean Average Precision per frame, but municipal asset managers need segment-level, georeferenced inventories. We argue that under BSI PAS 2161:2024 a segment-level severity score may align more closely with municipal decision-making than detection-mAP alone. We present an end-to-end pipeline from monocular dashcam imagery: SAM 3 produces both the road-surface mask and the per-detection instance mask; monocular depth (MoGe and Depth Anything 3, with Depth Anything V2 as a metric fall-back) feeds a RANSAC ground-plane fit; ray–plane inverse perspective mapping projects each detection as a full polyline, anchored laterally to PDOK BGT cadastral road edges; detections are clipped to a 5 m forward window per frame and aggregated into 25 m segments with explicit CROW-aligned weights. On a single Delfzijl case study we demonstrate empirically that segment granularity is itself a ratification dimension: the same physical road reads as 37 % red at 10 m and 67 % red at 100 m. Output formats are conceptually compatible with PAS 2161 sub-section reporting. Quantitative detection benchmarks (per-class mAP, RDD2020 cross-evaluation) and pipeline-component ablations are deferred to a companion technical report in preparation; this preprint focuses on the methodological position and its illustration.