The short answer
Every run is sealed: we fingerprint the conditions and the raw data when it ends, so any later edit shows. Provenance travels with it — interface, model, location, capture time, run index and a source receipt — so the same conditions can be measured again. Judgment is separated from collection: citation is verified against the source (deterministic), while mention and sentiment are judged on their own track with labeled confidence levels. Entity disambiguation keeps a same-name company from being counted as your brand. The measurement standards are published in the methodology.
Why this matters
When the party grading the work also collects on the result, the number can always improve. A report built that way rarely survives the first question in a board room.
Conditions are sealed at measurement time and handed over so the round can be run again. Citation is settled by address matching, mention is judged separately, and each carries its own confidence level.
Questions buyers read next
Where this answer comes from — Published pricesHow we measureTerms, defined