Prices are recorded from listings already passing through the server during ordinary browsing — no extra upstream request is ever made for this. Each group keeps a full price distribution, an observation count, and the first and last date it was seen. No group is published until it reaches a minimum number of observations.
The sample follows what visitors actually browse, so it is not a full market census. The median is derived from the price distribution and is therefore approximate; the counts and the "share under $X" figures are exact. Prices move hourly and listings sell fast.