Six stages on three pages, held here as text. You write a real CSS selector, it runs against a real parsed document, and you see exactly what came back. Nothing is fetched and nothing is installed. The pages are shaped like the practice sandbox, never like a vendor portal.
You are graded on what you got back, not on how you got it. A dozen selectors pull the same four titles, and finding a different one from the person next to you is not a mistake. What is compared is the list of values, in order.
Nothing cleared yet.
Every scraping tutorial works, because it was written against the page as it looked that day. Stage four changes the page under you. Nothing errors, nothing warns you, and a scheduled run would have written an empty column over a good one. That is the reason a scraper is the last route on the collection ladder and not the first.
When you go back to your own tool, the same five things still apply, whichever language you picked: take the value from the attribute rather than the visible text, anchor on something that carries meaning, do not catch the site furniture, follow the pages, and never let a dash become an empty cell that somebody reads as a zero.