Reviewing acquired datasets and third-party sources

Split the collection by origin

A combined dataset may contain first-party records, purchased feeds, licensed archives, scraped pages, commissioned examples, and synthetic content. Keep those categories separate. Their acquisition costs, available evidence, and permitted downstream uses may differ even when the final files share a schema.

For each source, record the original provider, acquisition method, date, applicable agreement or terms, and transformations. If a prior owner performed the collection, ask for its records rather than accepting a generic assurance that everything was public or proprietary.

Review the actual grant of rights

Check whether the original agreement allows sublicensing, redistribution, model training, commercial deployment, and access by the intended buyer. A right to use data in your own product is not the same as a right to sell a training corpus. Acquiring a business also requires review of which assets and rights transferred.

For scraped material, do not assume public availability resolves copyright, privacy, contractual, or access issues. For synthetic material, document the generation process and any source inputs or model terms that affect the proposed use. Get specialist advice where the rights position is uncertain.

Test provenance and quality independently

A documented license does not guarantee useful data, and high-quality examples do not prove clear rights. Check both. Sample across providers to find repeated content, mismatched schemas, stale labels, and inconsistent annotation standards. Keep source identifiers available for exclusions and corrections.

Ask whether multiple suppliers may have delivered the same underlying material. Report the method used to estimate overlap rather than presenting combined file counts as unique coverage. If a source cannot be traced well enough to answer a material buyer question, segregate it until the gap is resolved.

• Keep supplier restrictions attached to records or batches.
• Identify generated labels and synthetic examples.
• Track source-specific exclusions and quality limits.

Design a license around what is defensible

Offer the subset whose origin and permitted use you can support. Do not let a clean majority obscure a problematic minority. Explain excluded sources and any limitations they create for coverage. A buyer may prefer a smaller collection with inspectable evidence over a broader one with unresolved provenance.

Use source-aware delivery manifests and a process for later supplier corrections. If onward rights are too narrow for the intended buyer, negotiate an appropriate grant with the source or choose another collection. A new contract with your buyer cannot repair rights that were never obtained upstream.