Company data sources, provenance and evidence

Understand where company records come from, how data provenance supports a review and what a source-backed finding should contain.

What is data provenance in a company review?

Data provenance is the history that lets you trace a value or finding back to the record used to produce it. For a company review, that means more than a provider logo. A reviewer needs to identify the legal entity, the original record, the dates and any transformation between the source and the displayed value.

Data lineage describes how information moves and changes between systems. The two ideas meet when a registry record is collected, mapped to a company profile, compared with a previous snapshot and summarised in an investigation. Each step should preserve enough context to explain the result.

For our collection process and examples of the official registries behind the data, visit How We Source & Collect Data.

Which company data sources are relevant?

Source or recordUseful forQuestion to ask
Company registerEntity identifiers, registered details and available filing historyWhich jurisdiction and legal entity does it cover?
Shareholder filingRecorded holders, interests and available datesDoes it disclose the full chain or only one layer?
Corporate linkage datasetParent and group relationshipsHow does this source define a parent?
Financial statementReported figures and accounting contextWhat period, currency and reporting scope apply?
Collected snapshotReproducing the information used in a comparisonWhen was it retrieved and which source references remain?

the data provider documents company, director, shareholder, group-structure and financial fields in its data dictionary ↗. CompanyDelta’s coverage page presents selected field definitions and the supplied country statistics. Counts establish dataset scope; they do not prove every field is available for an individual entity.

An official record still needs context

A source can accurately report what was filed while the filing remains incomplete, delayed or inconsistent with other information. Keep the claim precise: “the available filing records this shareholder” is different from “the complete ownership position has been independently verified.”

As a UK example, Companies House’s register guide ↗ explains the public company information and filing records available through its service. Other jurisdictions publish different data. Do not extrapolate one register’s coverage or update process to every country.

Preserve three different dates

DateMeaningCommon mistake
Effective dateWhen the recorded event takes effect, if suppliedAssuming this is the retrieval date
Filing dateWhen a document or notice was filedTreating it as the date of every fact inside it
Detection or retrieval dateWhen your system observed or captured the informationCalling all retrieved values current as of that date

For financials, add the reporting period separately. A new filing can contain an older period or restated comparative figures. If a date is unavailable, state that directly rather than inferring it from the time the source was checked.

What belongs in an evidence packet?

  • Company name, country, registration number and internal reference.
  • The changed field and both the previous and updated values.
  • The exact records used in the comparison, with source references where available.
  • Available effective, filing, reporting and retrieval dates.
  • The method or calculation supporting the finding.
  • Unresolved gaps, conflicts and assumptions.
  • A responsible reviewer, proposed next step and recorded decision.

Retain the evidence used for the review even if later records change. A saved snapshot helps explain the decision at the time; a link to a source that has since changed may not. Avoid presenting a saved snapshot as an original filing when it is actually a collected or transformed record.

How should AI use the evidence?

AI can reduce the work of reading, comparing and summarising captured records. It should identify which evidence supports each finding and distinguish observed facts from interpretation. A plausible explanation without a supporting record is not a verified finding.

CompanyDelta’s connected investigation workflow prepares a draft from the available captured evidence for human review. The reviewer should inspect citations, calculations and missing information before accepting an outcome. The website’s sample briefings use clearly labelled invented records.

Questions to ask when evaluating a data provider

  1. Can we trace a displayed field to a source record, and what happens when no source link is available?
  2. How are company and person matches made, and how are uncertain matches represented?
  3. Do relationship labels distinguish shareholding, parentage and other connections?
  4. Are failed retrievals separate from missing disclosures?
  5. Can we preserve the evidence used in a decision under the agreed data rights?
  6. Can we test these points on our actual countries and entity types?

Use the company due diligence report template to keep these questions attached to the review. Evidence quality should be evaluated through a representative sample, rather than inferred from a headline coverage number.

Questions about sources & evidence

Does a source link guarantee the finding is correct?

No. The link helps you inspect the record. You still need the correct company match, compatible dates and a valid interpretation of the field or relationship.

Is a saved snapshot the same as an original document?

No. A snapshot preserves collected information. An original filing is the document submitted to or published by the source. Describe each accurately and retain both where available.

What should happen when sources disagree?

Preserve the conflicting values and their dates and definitions. Route the disagreement for review instead of silently selecting a value without an explanation.

Put the evidence to work.

Assess company coverage and build a review workflow around your portfolio.

Open your workspace
CompanyDelta