What the work actually involves

You are not manning a helpdesk. You are producing the material an AI model learns diagnostic reasoning from — which means writing out the parts of your job that usually stay in your head. A typical task looks like a user-reported symptom (VPN connects but internal DNS fails; a MacBook wakes from sleep without Wi-Fi; a Windows profile loads temporarily after a policy push) and asks you to lay out the hypothesis, the order you'd test in, the commands or GUI paths you'd use, what each result rules in or out, and the resolution. Other tasks flip it: a model produced a troubleshooting walkthrough and you judge whether it is correct, safe, ordered sensibly, and explainable to a non-technical user.

The common failure in submitted work is a correct answer with no visible reasoning. "Flush the DNS cache" is a step; "the host resolves by IP but not name, and nslookup against the internal resolver returns NXDOMAIN while the public resolver answers, which points at split-horizon DNS not being applied over the tunnel" is training data. Expect to write more prose per ticket than you ever did in production.

What the screen looks for

  • Real Tier 2/3 depth. Follow-ups go past the runbook: why that step, what would you check if it didn't work, what could this break.
  • Networking fundamentals you can explain, not recite. TCP/IP, DNS resolution order, VPN routing and split tunnelling, DHCP behaviour.
  • Cross-platform range. Windows is the baseline; comfort with macOS and Linux desktop issues widens the task pool.
  • Documentation habit. Ticket-system experience matters because it shows you already write for a reader who wasn't there.

Logistics

Contractor, fully remote, asynchronous — no on-call, no shift coverage, no queue SLA. You pick up batches and complete them on your own clock, with review feedback arriving after submission. Most contributors treat this as part-time alongside a day job; sustained availability of roughly 10–20 hours a week is what keeps a task pipeline open. Pay in the $30–55/hr band has been observed on this project and typically tracks task complexity and quality-review scores rather than being a flat guaranteed rate. No prior AI or annotation experience is required.