the basics like logging, backoff and validation are fine, but the failure that really hurts is a selector that confidently grabs the wrong element, like a price that picks up the old crossed out value. getting the same output every run shows the extractor is consistent but it's not necessarily correct.

empty fields can also be systematic, so it helps to check completeness by page type and category not just overall. proxies mostly help with ip limits and geo blocks. fingerprinting and behavior need their own tests