Who builds the URL list?
This is the fork in the road. Pricemancer is given a domain and works out the product pages itself, running every discovery method the site supports and merging the results: sitemaps, store-platform bulk endpoints, category walks and site search. A sitemap covering 80% of a catalog plus a category walk covering a different 70% together beat either alone, so they all run rather than stopping at the first that works.
What happens on a shop that fights back?
Some storefronts are genuinely defended. Pricemancer escalates on its own — a plain HTTP request, then a request profiled to look like a real browser, then full Chromium, then a hardened browser build for the worst cases. Worth asking of any tool: what does it do here, and do you pay for the failed attempts? Here you are charged for the technique that actually succeeded.
Does collection speed adapt, or is it a fixed guess?
A single global rate limit is either too slow for tolerant sites or too aggressive for fragile ones. The collector raises parallelism while throughput keeps improving and backs off when a site starts pushing back, so each shop runs about as fast as it will tolerate.
How much re-work does a site redesign cause?
Configuration written per shop is configuration that breaks when the shop changes. Because discovery and extraction are driven by what the page actually exposes rather than hand-written selectors, a redesign usually means re-running discovery, not rebuilding a mapping.
What happens to the pages that fail?
Some URLs are dead products, not blocked ones. Pages that fail repeatedly are flagged and skipped on later runs so they stop consuming time and tokens, and they stay visible in the interface so you can see exactly which ones rather than wondering why a count is short.