From real-world events to open POI data: a measurement protocol for time-to-reflection, with a benchmark from Japanese official registers

Abstract

Background Open point-of-interest (POI) datasets increasingly carry temporal fields—Foursquare OS Places attaches creation and closure dates to over 100 million places and feeds Overture Maps—inviting longitudinal research that depends on a quantity never systematically measured: the delay between a real-world event and its reflection in the data. Methods We decompose this time-to-reflection into incorporation lag and detection lag, estimated as survival curves with calendar-cohort stratification and Greenwood confidence intervals, plus a conditional field offset. Validity rests on protocol steps with measured costs of omission: three reference-qualification criteria (whose enforcement excluded our largest candidate source), terminal-closure filtering of references, an at-risk denominator, and linkage-error propagation. References—Japanese postal ledgers (two independent legal sources, cross-validated at 84% agreement with a 0-day median date difference), bank-consolidation notices, and chain press releases—are released alongside the paper with acquisition manifests. Results Incorporation accelerates across cohorts: the 90-day incorporation rate of chain openings rises about tenfold within the single brand observable across all cohorts (the pooled twentyfold contrast decomposes into acceleration, brand composition, and platform existence—about sevenfold once openings predating the platform, structural zeros, are excluded); post offices reach 63.6% incorporation within five years. Detection does not converge: 21.6% [17.9, 25.3] of final post-office closures are marked within five years in the pooled view, 24.3–33.3% across the cohorts with full five-year exposure (mature-subset coverage 33.8%; all upper bounds for the identifiable universe), and detection risk shifts with calendar period (log-rank p = 0.002). Ghost incorporation—entry after closure—affects 8.2% of date-complete matched final post-office closures (66/804). In the bank universe, a platform pre-history design—restricted to closures predating the platform’s founding, which structurally excludes denominator ambiguity—gives a lower bound of 1.4% (5/363); the two rates condition on different denominators and are not directly comparable. Conclusions Every category needs its own external clock: marking rates differ tenfold across categories, and no single lag constant is meaningful across cohorts. The protocol, reference event sets, and matched pairs enable any country or category with a qualifying register to build its own.