In one line

A miner's code is compared for originality, then built and run inside a judging image nobody outside RedTeam can read, and only that image decides whether the attack or the defense actually worked.

What this shows

Two live challenges share one payout: AB Sniffer (detect automation, weight 0.4) and Bot Virus (evade detection, weight 0.6).
A submission is compared for originality against every past accepted solution and everything submitted that same day, before it is ever run.
The actual judge — the code deciding pass or fail — ships only as a versioned Docker image pulled from a private registry. The public repository declares its scoring methods but never implements them.
A passing score decays over 10–15 days; standing still costs a miner its place.

Read this next to the lesson

Every box below links to the exact quote or code path it is drawn from — hover or click a box, then use How it's built for the file and line.