What the work actually is

You build network problems and then judge how well a model solves them. A task might start as a design question — dual-homed BGP with a preferred transit and a backup, redistribute or don't, what happens to the routing table when the primary session flaps — or as an outage narrative where a VLAN mismatch, an asymmetric firewall path, or an MTU problem on a GRE tunnel is producing symptoms that look like something else. You write the scenario with enough topology detail that it's solvable, write the reference answer, and write out the reasoning: why this design decision and not the plausible alternative, which show command narrows the fault domain and why.

The grading half is where the judgment lives. Models produce configurations that parse cleanly, use real syntax, and would still take down a production network — a route-map that leaks more than intended, an OSPF area design that breaks on the second failure, an ACL applied in the wrong direction. Flagging those, and explaining precisely what breaks and under what conditions, is the part the labs are paying for.

What the screen looks for

  • Production exposure, not lab-only familiarity. Screeners probe for what you did during a real outage, what the initial hypothesis was, and how it turned out to be wrong.
  • Protocol depth that survives a follow-up. Naming BGP attributes is table stakes; explaining why local-pref beats AS-path prepending for a specific inbound-versus-outbound problem is the actual signal.
  • Written reasoning. Reference answers are documents other engineers and graders rely on, so the writing sample is effectively part of the qualification.
  • Calibration in grading — being able to say an answer is technically correct but operationally reckless, and defend the distinction.

Logistics

Fully remote and fully async: no standups, no shared time zone requirement. Work is assigned as ongoing project batches and you set your hours each week, with a stated floor of 10. Pay is observed at $80–120/hr and disbursed weekly via Stripe; rate placement generally tracks demonstrated depth and the complexity tier of the tasks you're cleared for, and is not guaranteed. Expect an initial calibration period where your scenarios and grades are reviewed against other reviewers before you're routed higher-difficulty work.