Ask a bounded question
Define what the research must decide, which public sources are in scope and the fields to collect. A useful source record names its publisher, URL, retrieval date and coverage. A search result or an extracted sentence is a candidate until its context has been checked.
Keep disagreement visible
Different sources may use different dates, currencies, entities or counting rules. Preserve the original definitions before merging rows. Deduplication should not erase conflicting evidence. Mark unavailable fields as missing rather than substituting zero. These simple choices make downstream review and reuse much easier.
Build a small sample before a spider
Collect a small sample manually and verify whether it supports the decision. Then define the repeatable extraction and review steps. This follows LOBE: reuse an existing source or tool before building another crawler. The integration offer starts from an agreed source inventory and output schema, not a promise to cover the whole web.