Cluster deduplication threshold
Track-B Phase B: site-selection redaction batch 2 — redact exact clustering/dedup/proximity thresholds and named production dataset counts (R1) from cluster-formation, cluster-deduplication-threshold, co-location-intelligence-overview, asset-evaluation-protocol, co-location-investment-thesis, co-location-strategy, co-location-ranking-system; generalize to structural-fact level, no fabricated substitutes
@@ -11,9 +11,9 @@ status: active audience: customer-woodfine bcsc_class: current-fact language_protocol: PROSE-TOPIC last_edited: 2026-08-25 last_edited: 2026-09-04 editor: pointsav-engineering short_description: "The co-location index pipeline deduplicates overlapping clusters that represent the same commercial zone using a fixed, tightly-set proximity threshold, retaining the cluster with the higher secondary operator count. An earlier, substantially wider threshold was found to suppress legitimately distinct commercial nodes in dense suburban corridors." short_description: "The co-location index pipeline deduplicates overlapping clusters that represent the same commercial zone using a fixed, tightly-set proximity threshold, retaining the cluster with the higher secondary operator count." paired_with: site-selection/cluster-deduplication-threshold.es.md cites: - ni-51-102 @@ -24,21 +24,21 @@ The [[co-location-methodology|co-location index]] pipeline produces one cluster ## The same-parking-lot problem Large commercial zones frequently host two or more anchor-category stores within metres of each other. A Home Depot and a Costco sharing a parking lot in suburban Edmonton, for example, sit roughly 20 metres apart. Without deduplication, both stores produce clusters with nearly identical catchment geometry, co-tenants, and scores. The map displays two concentric rings covering the same zone — neither wrong in isolation, but together misleading about the number of distinct commercial nodes in that corridor. Large commercial zones frequently host two or more anchor-category stores within metres of each other — a home-improvement superstore and a warehouse club sharing a single parking lot is the common case. Without deduplication, both stores produce clusters with nearly identical catchment geometry, co-tenants, and scores. The map displays two concentric rings covering the same zone — neither wrong in isolation, but together misleading about the number of distinct commercial nodes in that corridor. ## Threshold selection The deduplication step removes any cluster whose anchor store falls within a fixed, tightly-set radius of a higher-ranked cluster anchor already confirmed for retention. The radius is narrow enough that only stores genuinely sharing a single parking lot or immediate building complex are collapsed. Anchors in adjacent strip malls separated by a service road are treated as distinct nodes and retained. An earlier implementation used a substantially wider threshold. Field review of the Edmonton metropolitan area identified cases where legitimate, separately operated commercial zones were being suppressed without notification: a Walmart-anchored node and a Home Depot-anchored node in neighbouring commercial blocks, serving different residential catchments, were treated as duplicates and one was removed. That wider threshold proved too coarse for dense suburban corridors in Canadian and North American markets, and it was tightened accordingly. The calibration is deliberately conservative in that direction. A threshold set too wide collapses separately operated commercial blocks — blocks that serve different residential catchments — into a single node, understating the number of distinct sites in a corridor. Suppressing a legitimate node is the more damaging error of the two, because it removes a site from consideration without surfacing that anything was removed. ## Ranking the survivor When two anchors fall within the threshold distance, the retained cluster is the one with the higher count of co-tenants within the 3 km catchment radius. Ties break on the 1 km count. This ensures that the cluster representing the fuller commercial zone — more stores, broader multi-purpose draw — survives, regardless of which anchor happened to be processed first. When two anchors fall within the threshold distance, the retained cluster is the one with the higher count of co-tenants inside its catchment. Where that count ties, the tiebreak falls to the co-tenant count inside a tighter inner radius. This ensures that the cluster representing the fuller commercial zone — more stores, broader multi-purpose draw — survives, regardless of which anchor happened to be processed first. ## Pipeline effect After applying the tightened threshold, deduplication removes a meaningful share of candidate clusters as same-zone duplicates in a representative pipeline run. The reduction is concentrated in dense commercial corridors where multiple anchor formats co-locate in close proximity. Tier distribution and national rankings are assigned after deduplication runs, so the published counts reflect deduplicated zones only. Current production counts are published in [[about-regional-markets-system|Regional Markets Intelligence System]]. Deduplication removes a meaningful share of candidate clusters as same-zone duplicates in a representative pipeline run. The reduction is concentrated in dense commercial corridors where multiple anchor formats co-locate in close proximity. Tier distribution and national rankings are assigned after deduplication runs, so the published counts reflect deduplicated zones only. Current production counts are published in [[about-regional-markets-system|Regional Markets Intelligence System]]. ## See also