How to check a data pack before you trust it
You found a data pack. The listing shows a clean table, a row count, a price. Before that file touches your spreadsheet or your script, it deserves ten minutes of checking, because a pack that looks finished can still fall apart on the first question you ask it.
This checklist works on any data pack from any seller. I will use our World Civic Amenities Pack as the worked example, since every point below maps to a decision we made while packaging it.
1. Does it state how each number was counted?
A count with no method is a rumor. Look for the tool and the date. The amenities pack states that every figure was counted from OpenStreetMap on 2026-09-10 via the Overpass API, one query per city per amenity. If a seller cannot name a source and a snapshot date, you cannot reproduce anything they sold you.
2. Does it define the boundaries?
The same city can be measured several ways. A bounding box grabs outskirts an administrative boundary leaves out, and the gap is not small. The Mumbai figures come to 67 drinking water features inside the district boundary and 71 in a bounding box over the same ground, because the box clips the harbor differently. The six tables that ship with method notes are area-filtered on administrative boundaries, and each note gives the relation ID per city, so you can rerun the query yourself. One caveat is spelled out: the Melbourne numbers cover the City of Melbourne local government area, about 37 square kilometers of core municipality, not greater Melbourne. If a pack never says what area a row covers, treat that row as decoration.
3. Can you spot-check it?
The method notes list a real named feature with coordinates behind the counts, so you can open the map and confirm the numbers are not invented. Zeros deserve a look too. A zero can be a real measurement, and this pack says outright that no public bookcase is mapped anywhere in Lagos, Nairobi, or Addis Ababa, and that the Jakarta bookcase row has no node to show because the count is 0. A pack that explains its zeros is a pack that actually counted.
4. Did the seller run the numbers twice?
Collection services have mirrors, and mirrors sit at different snapshot depths. Three of the six method notes list a second Overpass count: all 9 figures identical on the Oceania and US-2 tables, and 8 of 9 on the Africa table, where the staler mirror put Nairobi drinking water at 2,173 against the 2,174 published, one edit in the gap between the two bases. One pass is hope. Two passes that agree are evidence, and you want to see those numbers written down.
5. Is there a checksum file?
Not in this one. The pack ships no SHA256SUMS.txt, so the honesty test is what the README does instead: it names every CSV in the zip and gives its row count, and the eight regional counts add up to the 84 rows in the master table. Check the file list against the archive you extracted before you open anything. A checksum file is the gold standard, and it costs the seller one command.
6. What does the license require?
Open data carries obligations. These tables come from OpenStreetMap under ODbL 1.0, which requires attribution and applies share-alike to derived databases. A pack that buries its license is handing you a legal problem later.
7. Is the seller honest about free alternatives?
The README of this pack says plainly that every regional table is already free to download, one table at a time, on the data-vault catalog. The paid pack is the combined work: one master schema across 8 regional tables, a 25-city summary, the method docs, and a frozen snapshot. A seller who names the free version of their own product is telling you the paid version has a specific job.
Run this list before your next purchase. The [worked example](https://www.getly.store/product/world-civic-amenities-pack-osm-counts-25-cities-8-tables-mtw1q9js) ships 84 rows across 8 tables and 25 cities, and its regional tables stay free either way. Seven answers in, you know whether a pack was built or assembled. If the seller skips the method, the boundaries, and the license, skip the pack.