Silence as Data: A Lambda-Architecture Pipeline for Locating Digitally Invisible Communities on a 1 km² Grid of Morocco.

Progress toward the Sustainable Development Goals (SDGs) is monitored from nationallyreported statistics, which silently exclude populations that leave no digital or administrative trace. Thispaper describes a prototype that treats the absence of such traces as a measurable signal. We harmo‐nize raster night-light imagery (GeoTIFF) and tabular indicators from UN, World Bank, FAO, ILO,UNDP and Moroccan HCP sources (CSV) onto a unified grid of 450,000 cells of 1 km², and derive sixaccess indicators per cell: electricity, water, hospital proximity, school proximity, financial inclusionand digital access. A Digital Presence Score on [0, 100], built as a weighted linear combination andchecked with principal component analysis (PCA), is segmented by K-Means into three connectivitytiers; DBSCAN then extracts contiguous low-score clusters, which suits the irregular shape of moun‐tain settlements. The pipeline follows a Lambda architecture: Python producers feed a four-topicApache Kafka broker, a Spark Structured Streaming speed layer writes to a Redis cache with a sub50 ms latency target, a Spark batch layer persists results in PostgreSQL, and a Node.js API serves aLeaflet/Chart.js dashboard, deployed as five Docker Compose services. The project documentationfixes these design parameters but reports no ground-truth validation, tier populations or measured la‐tencies. We state this as an explicit evaluation gap and propose a protocol to close it. We also arguethat, without a population mask, “shadow” cells conflate uninhabited terrain with underserved settle‐ments, and we identify this as the main threat to the approach.