Summary of an article published on The Web Scraping Club on 2026-08-25 (free to read). Read the full article

hermes.com sits behind DataDome, and a headless browser there pulls the whole product page every time: HTML, scripts, fonts, images, every XHR, all through residential proxies. The catch is that the page is only a wrapper. The frontend gets its product data from a JSON API on bck.hermes.com. Coming from sneakerbotting, we know that a request-based bot lives or dies on three signals: TLS fingerprint, header order and a valid antibot token.

We built a Go scraper around bogdanfinn/tls-client and fhttp, with the Hyper Solutions API minting the DataDome sensor data. It warms a session once, then calls the API directly and reuses the session. We found the catalog through the category tree, ran the whole US catalog on 8 concurrent sessions, and measured bandwidth and sensor calls per product. Then we priced it against buying a ready-made dataset, and projected it across markets.

What’s inside

  • Why header order has to match a real browser capture per host: the API calls and the DataDome captcha-delivery endpoints use different HTTP versions and header casing.
  • How the session is warmed: the DataDome challenge types handled, the tag posts, and the call that mints the XSRF token the API expects.
  • Why the category tree beats the sitemap for discovery, and how far the sitemap overstates the real catalog.
  • Which part of the cost is metered, and where buying the dataset still makes sense.

In the full article: The Go header block with explicit ordering, the session bootstrap diagram, the measured run metrics for the US catalog, per-market catalog sizes, the build versus buy cost figures and the repo commands.

Read the full article on The Web Scraping Club