Co-location Tier Nomenclature — PointSav Documentation

The co-location platform classifies every identified retail cluster into one of three tiers: T1 Regional, T2 District, and T3 Local. This article defines what each tier means, explains how the classification is determined, and documents the evolution of the naming convention.

[edit]The Three Tiers

T1 Regional denotes the highest-order co-location clusters: those anchored by the largest retail formats in the market. A cluster reaches T1 when it contains at least one warehouse club — a membership-based bulk retailer such as Costco, Sam's Club, or PriceSmart — or when it contains a full-format hypermarket accompanied by a hardware anchor. The hypermarket-plus-hardware combination is required for T1 designation because a hypermarket without a hardware complement indicates a smaller retail catchment than the combined offer. As of Phase 23+Change B, the platform identifies 1,746 T1 clusters across 18 countries.

T2 District denotes clusters that meet the hypermarket anchor threshold but do not qualify for T1. This encompasses two distinct situations. A cluster that contains a hypermarket and hardware anchor but whose member locations span more than 2.5 km from the geometric centre to the most distant member is classified as T2 rather than T1. The geometric span gate — introduced in Change B — reflects the finding that very widely dispersed clusters, even when they contain equivalent anchor composition to T1 clusters, represent different consumer catchment dynamics. Tighter clusters indicate denser co-location effects. A cluster that contains a hypermarket but lacks a hardware anchor is also classified as T2, regardless of its geometric span. The platform currently identifies 2,726 T2 clusters.

T3 Local denotes clusters that contain significant retail co-location but lack anchor formats sufficient for T1 or T2 designation. T3 clusters typically contain a mix of mid-format grocery, electronics, and lifestyle retailers without a full hypermarket presence. They represent locally important retail nodes rather than the regional and district drawing-power markets of T1 and T2. The platform currently identifies 2,021 T3 clusters.

[edit]The Two-Pass DBSCAN Algorithm

Tier classification is not assigned independently of cluster discovery: the tier emerges from the geometry and composition of clusters identified by the two-pass DBSCAN algorithm.

The first pass uses anchor stores as seeds. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) groups points that are within a defined radius of each other into clusters. In the first pass, only the anchor-category stores — warehouse clubs and full hypermarkets — are used as seed points. This pass identifies the cores around which high-order retail clusters form.

The second pass fills each identified core cluster with additional retail locations within the cluster's geometric reach. Hardware anchors, lifestyle retailers, electronics stores, and other category members are added when they fall within the radius of an existing seed cluster. The composition of the filled cluster — which anchor categories are present — determines the tier.

The geometric span measurement used in the Change B gate is taken after the second pass, when the cluster's full member set is known. Span is measured as the maximum distance between any two member locations within the cluster.

[edit]The Span Gate: Change B

The 2.5 km span gate was introduced in Change B (commit 84b7fe7a, 2026-05-28). Prior to Change B, tier was determined solely by anchor composition: if a cluster contained the right anchor mix, it was T1, regardless of how spread out its members were. The gate corrected for a finding that a subset of T2-composition clusters had span values well above 2.5 km and were representing geographically diffuse retail distributions rather than the tightly co-located patterns that define the market analysis use case.

The Change B span gate demoted approximately 667 clusters from T2 to T3, shifting the T2 count from approximately 3,393 (Phase 22) to 2,726 (Phase 23+Change B) and increasing T3 from approximately 1,354 to 2,021.

[edit]Naming History

The T1/T2/T3 nomenclature was adopted to provide a stable, non-hierarchical labelling system that conveys market function without implying a simple rank ordering. Earlier internal development used placeholder labels — Alpha, Beta, and Gamma — which were retired before any public output was produced. The T1/T2/T3 labels have been canonical since Phase 18 and appear in all published data files, TOPIC articles, and research manuscripts.

The descriptors — Regional, District, Local — are intended to convey catchment geography, not administrative hierarchy. A T1 Regional cluster typically serves a regional catchment extending 20–50 km; a T2 District cluster a more localised catchment; a T3 Local cluster a neighbourhood or small-town catchment. These are empirical generalisations, not definitional constraints.

[edit]Counts by Geography (Phase 23+Change B)

The platform covers 6,493 clusters across 18 countries. T1 clusters are most common in the United States (889), Germany (227), and France (247). T2 clusters are most common in the United States (1,779), Germany (338), and Great Britain (400). The full per-country breakdown appears in the co-location summary at gis.woodfinegroup.com/research-summary.html.

Data provenance: clusters-meta.json Phase 23+Change B (2026-05-29T05:00Z rebuild); taxonomy definitions in app-orchestration-gis/taxonomy.py and config.py.