What the work actually looks like

You sit inside an annotation platform for most of a shift. A task arrives — a prompt and one or more model responses, or a search result set — and you apply a project-specific rubric to rate helpfulness, factual accuracy, instruction-following, and safety. Where two responses compete, you rank them and write the justification that explains the gap. Where a response is confidently wrong, you locate the error, cite what the correct answer is, and log it in the format the project expects. Alongside grading, you author prompts designed to stress the model: edge cases, ambiguous intent, multi-step reasoning, requests that invite a plausible-sounding hallucination.

The less visible half of the job is consistency. Rubrics are written in English, applied by dozens of people, and ambiguous at the margins. You will be asked to re-review your own past ratings when they drift from the guideline, to flag genuine gaps in the rubric rather than silently improvising, and to argue your position in a reviewer thread when your score is overturned. Quality reviews are ongoing and scores follow you.

What the screen looks for

micro1's process is AI-led and front-loaded. Expect a structured interview that probes your prior annotation or evaluation work in specifics — which platform, which rubric dimensions, what your agreement rate was, how disputes were resolved — followed by follow-ups that test whether you actually did the work or watched someone else do it. Generalist background is fine; the platform is screening for people who can read a guideline closely, apply it the same way on task 400 as on task 4, and explain a judgment call in writing.

Logistics

  • Contractor, fully remote, asynchronous within project deadlines.
  • Open to candidates based in North America and Europe.
  • Work volume fluctuates with client demand; treat it as variable rather than a guaranteed weekly load.
  • Pay is observed in the $14–36/hr band — the spread reflects project type and demonstrated evaluation experience, and is not guaranteed for any individual engagement.