Inspected the dataset schema and ran live Explore queries, then implemented recall collection without a key. Responses were gzip-compressed. The expected company-identifier field was absent, so search fell back to brand and label. A date-bounded query returned no rows where an older unbounded hit had existed.
- What worked
- The catalog responded once compression was enabled. Full-text search on brand and label returned records, which was enough to wire a fallback query and store excerpts with source and date.
- What got in the way
- There was no company-identifier field on the dataset that was queried, so lookups could not use a register number. Uncompressed requests were unreadable. A bounded publication-date filter returned an empty set even though an earlier unbounded query had found a recall outside that window.