Work

When Splitting Runs Out: Diagnosing and Bounding Result-Cap Truncation in a Platform-Derived National Frame

Year: 2026
Type: preprint
Abstract: Map platforms cap results per query; splitting assumes every capped query can be split below the cap. We examine a Google Maps frame of tattoo studios in Turkey: 4,854 open businesses in 429 districts. Its log records 1,121 capped, unsplit queries, whose results flag nine districts where truncation is likely. Splitting 295 neighbourhood queries into 1,613 street queries recovered 0.24 businesses per parent query when the district's own studios filled at least 70% of the parent list, against 0.04 otherwise. In the nine flagged districts, a Chao2 lower bound puts missing listings at 0.52% of the frame, and an OpenStreetMap audit found all 51 open, identifiable studios in the frame, a one-sided 95% upper bound of 5.7%. Two recovery routes failed, and a photograph selector drawing in other businesses' images made a vision-language model read another shop's sign in 268 of 1,229 audited sign-present decisions. A reporting checklist follows. Data and code availability: Anonymized code and aggregate outputs are provided as supplementary material (plain text). Listing-level data are not shared because they contain business names, addresses and telephone numbers.
Language: English
Open Access status: green
OpenAlex ID: W7233553270
Cites:
Cited by:
Related to: