Roman Sorokin

I build scrapers with one rule: an empty field beats a wrong value.

Most scrapers hand you a number and let you assume the rest. Amazon returns prices in the currency of whatever country the request landed in, and says so nowhere in the markup — so 390037 could be dong for a mouse or dollars for a machine. My actors put the ISO code next to every price, and flag the record when the currency is not the one that marketplace should have returned.

The same rule everywhere else:

• A missing price is null, never a number borrowed from elsewhere on the page.
• A block is reported as a block. Anti-bot pages arrive under HTTP 200 in about 2 KB — treating that as success is how a scraper reports "fine" while returning nothing at all.
• Numbers are parsed per locale. 1,234.56 in the US and 1.234,56 in Germany are the same amount; 8,63 in Poland is eight units, not eight hundred. Stripping commas is wrong by a factor of 100, and wrong silently.

Amazon International Scraper — 18 marketplaces, verified on every one of them.