Summary of an article published on The Web Scraping Club on 2026-09-10 (for paid subscribers). Read the full article
Anyone scraping a DataDome site knows the theory: the tag reads the browser fingerprint, watches how you use the page, and a server turns that into a score. The target is leboncoin.fr, running DataDome JavaScript Tag 5.9.4. After the Kasada work on nike.com in THE LAB #114, I expected a similar interpreter, and the sensor turned out to be built differently, so I compare the two products at the end.
We used Manus for the first pass on the code and Claude to check every claim against the bytes and the wire, with read-only Camoufox captures from a residential connection in Italy. I cover how the five DataDome pieces are installed on the site, what the sensor posts, and how the fingerprint probes group into families. Every claim carries an evidence level, so a string in a table is never mistaken for a property that was sent.
What’s inside
- When you say “DataDome blocked me”, you should know which of the five pieces you mean: the remote component, the sensor, the collection endpoint, the two bootstraps or the page the user sees.
- The evidence levels are TABLE, SOURCE and WIRE, and the difference between a string that exists, code that can run it and a value actually seen in a request is most of the work.
- Manus reviewed its own report with three blind audits and an adversarial pass, and lowered claims the evidence could not support.
- Server-side scoring is out of scope because nobody outside DataDome can see it, and the sensor also sends part of the site configuration back to DataDome on a sampled basis.
In the full article: The full list of fields in the sensor POST, the fingerprint probe families, the Manus outcome matrix, the claim table in The Lab repository and the side-by-side comparison with Kasada.