How we build
reference editions.
Every edition ships with a documented chain of custody. Seven steps, applied to every product line, every edition, every release.
STEP 01
Source policy.
Every edition is sourced from license-clean upstream datasets. Source families are selected per product line and recorded in the Provenance and License Notes inside the pack. Sources with share-alike obligations that conflict with commercial redistribution are excluded. Standalone API-driven sources whose terms forbid resale are excluded.
For F&B Local Discovery, the 2026 source is Overture Maps Places under CDLA-Permissive-2.0, with Foursquare-derived rows entering through Overture under Apache-2.0 with NOTICE handling preserved.
STEP 02
Curation.
Inclusion logic uses primary category signals only. Alternate-category and hierarchy fields are not used for inclusion because they pull in places whose primary identity does not match the buyer's question. This is the difference between "places that mention restaurants somewhere in their tags" and "places that are actually restaurants."
STEP 03
Geography filter.
Geography is enforced at the country level using the upstream address attestation, with a bounding-box fallback for rows that lack a parsed country. United States editions are filtered to United States only. International editions get their own country-level filter.
STEP 04
Deduplication.
Three tiers, conservative by design:
- Tier 1. Exact brand identifier plus tight coordinate match. Auto-merge.
- Tier 2. Exact name plus tight coordinate match. Auto-merge.
- Tier 3. Fuzzy name similarity plus loose coordinate match plus matching category, with a strict confidence floor. Auto-merge above the floor; flagged for analyst review below.
Records that fail the confidence floor remain separate and are tagged with a Dedupe Review Flag column so analysts can find them. The pack does not silently merge records below the threshold.
STEP 05
Documentation.
Every edition ships with seven designed PDFs covering README, Start Here, Data Dictionary, QA Report, Provenance and License Notes, Agency Use Recipes, and Limitations and Non-Advice Notice. Documentation is written buyer-facing in plain English, not in pipeline jargon.
STEP 06
Attribution.
Each edition includes the upstream attribution text and the relevant NOTICE files, packaged so buyers can redistribute derivative work in compliance. The Provenance and License Notes document the obligations that apply.
STEP 07
Reproducibility.
Every edition records the upstream release identifier, the retrieval timestamp, and the build version in a provenance manifest. SHA-256 hashes accompany the final ZIP so buyers can verify integrity. Editions are dated and stable; new releases are issued as new editions, not as silent overwrites.