When an enrichment fills a field with something wrong, the useful question is not what it wrote but why. Without a record, tuning an enrichment is guesswork with a bill attached.
Every automation run is now logged per lead, and the log is one click from the lead. It shows which enrichment ran, the inputs it used, the raw response it got back, the fields it wrote, and the connects it consumed. AI enrichments also record the model's own explanation of its answer.
JSON responses render as expandable tables rather than a wall of text, so you can actually read what a vendor API returned.