ScrapeSentry
Judge demo

Three minutes, one continuous recovery.

A real Bright Data collector breaks the way real sites break. Watch the contract catch it, hold every unsafe row, and release new data only after a human approves and both verification gates pass.

Try it yourselfView source
2 minutes 55 seconds. Narrated, no cuts through the recovery flow.

What happens, and when

  1. 0:00
    About the projectWhy a 200 OK response is not proof that the data is safe.
  2. 0:27
    Tech stack and architectureOne typed service layer behind FastAPI and FastMCP.
  3. 0:59
    The fleetOne real Bright Data collector on contract v1.
  4. 1:11
    The silent breakprice becomes price_text, availability becomes stock_label.
  5. 1:23
    DetectionHealth 100 to 64, twelve rows quarantined in 47 ms.
  6. 1:35
    Downstream stays cleanConsumers keep the last verified rows, flagged stale.
  7. 1:50
    Self-Healing and the diffThe candidate is contract tested before approval.
  8. 2:12
    Approve, then verify twiceCanary and production gates against the same contract.
  9. 2:28
    Eighteen MCP toolsThe same incident, resolvable from a coding agent.
  10. 2:38
    What I learnedProving the repair is safe is the hard half.

What is recorded and what is live

The public deployment runs in demo mode against a captured Bright Data production run, so judging costs no provider credits. The provider evidence panel says so on screen rather than implying a live call.

The underlying run is real: collector c_msxcshnz2bwrroekf3, collection job j_msyad7l21v1fnqbljn with twelve records across thirteen pages and zero failed crawls, and Self-Healing job ia_msya56gd1y2ucwwy0j, which published production Version 2. Every artifact is committed in the repository.

Inspect the provider evidence