Combining two public sources can add context and also create new assumptions

Adding a column from a second public source often feels like a straightforward upgrade. A directory gains corporate status. A list of locations gains industry information. A procurement file gains website details.

The new field may be useful. But it is not merely extra information. It is a claim borrowed from one record and placed beside another. Before using that claim in a filter, ranking, count, or decision, name the assumptions that made the transfer seem reasonable.

A match answers one question: these records probably refer to the same entity. It does not automatically answer another: does this incoming value properly describe this particular row?

That distinction matters because rows, sources, and fields often operate at different levels. A local directory row may represent one storefront. A corporate filing may represent the legal entity that owns several storefronts. The name-and-address match can be sound while the imported status is too broad to label the storefront as operating, active, or closed.

Treat each imported field as a borrowed claim

A compact pre-join review can prevent much of the false certainty that follows enrichment. For each new field, ask five questions.

Is it the same entity? Identify what makes the match persuasive: a stable identifier, a name and address, or several matching details. Names alone can be weak evidence, especially where common names, brands, parents, and subsidiaries overlap.

Is it the same unit? Write down what one row represents in each source. One row might be a company, location, legal entity, brand, product, or reporting period. A company-level revenue value does not become branch-level revenue because it has been copied onto branch rows.

Does the field have the same definition? Labels that look familiar can conceal different concepts. “Active” might mean registered with a government agency, listed in a directory, licensed, recently observed, or believed to be operating. These are related facts, not interchangeable ones.

Is it for the same time? A source publication date is not necessarily the date each value was observed. A current website, an annual filing, and a historical directory record can all be accurate while describing different moments. Time belongs in the join condition, not just in a footnote.

What does no match mean? “No match found in this source” is not the same as “does not exist.” The source may have incomplete coverage, the organization may use a different name, or the matching rule may be deliberately cautious. Ambiguous and unmatched rows should remain explicit states.

These questions are not a guarantee that a linkage is correct. They make its reasoning visible, which is often the difference between useful context and an overconfident conclusion.

Identity and applicability are separate tests

Consider a local-business directory joined to a state corporate registry by name and address. The directory row identifies a storefront; the registry entry identifies a legal entity. The registry’s filing status can be valuable context, provided it is labeled that way.

Trouble begins when that status is silently reinterpreted as a location-level operating status. A business can have an active registration while a particular location has moved, closed, or changed hands. The match may be correct; the application of the attribute may not be.

The same issue appears in one-to-many relationships. Suppose a list of branch offices is linked to a company-level revenue source. Copying the parent’s revenue to every branch changes the meaning of the measure. Summing it afterward can multiply the same company value several times.

Before aggregating, check whether one record from either source can attach to multiple records in the other. Then decide whether the value should be held at its original level, allocated by a defensible rule, or excluded from the calculation.

A high match rate is not proof of quality

It is tempting to report that 90 percent of records matched and treat that as evidence that the work succeeded. Match rate measures how many links were made, not whether those links are right. The UK Office for National Statistics cautions that match rates do not indicate linkage quality; both incorrect links and missed legitimate links need attention.

Cautious matching can leave valid records unmatched. Loose matching can create an impressive match rate by attaching the wrong records. Neither outcome is understood by the percentage alone.

A practical check is to review a small, purposeful sample: accepted matches, rejected candidates, and no-match cases. Look for recurring causes of error, such as parent-versus-location confusion, stale addresses, shared phone numbers, or names reused by unrelated organizations. The review need not claim a precise accuracy rate to expose assumptions hidden by a clean-looking table.

Keep enough context to revisit the decision

For each enriched field, retain a short assumption card in the spreadsheet, database, or project notes:

• imported field and original source value
• source name and source date, plus observation date when available
• what the destination row represents
• what the imported value represents
• match basis and relationship cardinality
• reference date and transformation applied
• meaning of no-match or ambiguous status
• review status

This is modest documentation, not bureaucracy. The U.S. Census Bureau’s statistical quality standards call for defined linkage objectives and criteria, testing, evaluation, and documentation. The same discipline scales down well for ordinary business research.

It also supports more careful language. Rather than saying, “This list shows 800 active companies,” say, “This list contains 800 directory records linked to a corporate-registry record marked active as of the registry’s stated date.” The second statement identifies the list, the link, and the scope of the status.

Combining public sources can make a dataset far more informative. It can also introduce a chain of assumptions that neither source stated on its own. Name those assumptions before adding the column, and the resulting analysis will be easier to defend, update, and use responsibly.