Regional market definition
docs(site-selection): reconcile co-location-tiering-scoring into the canonical four-tier article -- the T1/T2/T3 points-based compositional scale it documented was retired platform-wide 2026-05-16 and describes no current cluster. Merged its two passages with no equivalent anywhere else (the cluster-count parameter-sensitivity finding, and the full planned per-cluster strength score: design principles, three-driver set, open weighting question, scorecard contents) into catchment-ranking-methodology-v3; did not carry over the retired T1/T2/T3 material or the cluster-formation recap already covered by co-location-cluster-formation and cluster-deduplication-threshold. Source archived with superseded_by, redirect added, 8 inbound references repointed. Also removed an ES-only table row in regional-market-definition.es.md carrying DBSCAN eps/minPts/IoU parameters and a 226-476 sweep range that the EN pair had already had redacted -- restores EN/ES parity
@@ -37,7 +37,7 @@ Two failure modes follow from conflating coverage with market: **Single-anchor settlements dominate the population.** A town with one qualifying co-location is, under the current rule, the same object class as a metro holding many. The label "Regional Market" implies concentrated retail demand; the rule admits the single-anchor case. The headline count is therefore inflated by exactly the cases that carry the least decision value for site selection or investment. **The count reads as an artefact of the floor, not the geography.** A reviewer can move the count up or down simply by arguing the floor, which is the classic sign that the threshold — not the data — is doing the work. This is the same failure class as the DBSCAN parameter sensitivity documented in [[co-location-tiering-scoring|the co-location tiering and scoring methodology]], where parameter sweeps move the North American cluster count across a wide range without any change to the underlying retailer data. **The count reads as an artefact of the floor, not the geography.** A reviewer can move the count up or down simply by arguing the floor, which is the classic sign that the threshold — not the data — is doing the work. This is the same failure class as the clustering parameter sensitivity documented in [[catchment-ranking-methodology-v3|the V3 catchment ranking methodology]], where parameter sweeps move the North American cluster count across a wide range without any change to the underlying retailer data. The geocoding-confidence field does not resolve this. It is geocoding precision — specifically the quality of the boundary assignment — not market quality. It must not be presented as a ranking or a quality signal. @@ -100,6 +100,6 @@ Two honesty notes belong in the platform's documentation alongside this table: ## See also - [[about-regional-markets-system|Regional Markets Intelligence System]] — the full dataset, tier system, and qualification method this article's definitions build on - [[co-location-tiering-scoring]] — how tiers and the planned strength score are computed for each co-location inside a Regional Market - [[catchment-ranking-methodology-v3]] — how tiers and the planned strength score are computed for each co-location inside a Regional Market - [[trade-area-methodology]] — how the trade area for each co-location is defined - [[spend-population-provenance]] — the estimation chain for population and spend figures attributed to each co-location