What the work involves

This is documentation work pointed at model training rather than at end users. On a typical task you'll be handed a code sample, a spec fragment, or a half-broken internal doc and asked to produce the reference-quality version — or to judge which of two model-generated versions is closer to it and say precisely why. Assignments cluster into a few shapes: authoring gold-standard reference answers for prompts a model failed, rewriting model output to publishable standard with an annotated diff, scoring pairs of drafts against a rubric covering accuracy, structure, and audience fit, and flagging documentation that is fluent but factually wrong — the failure mode that matters most.

The work rewards people who have actually shipped docs for a real product. You will be asked to reason about whether a code sample compiles, whether a parameter table matches the signature, whether a procedure's steps are in an order a reader can follow without backtracking, and whether the doc is aimed at the right reader. Generic prose polish is not what's being bought.

What the platform screens for

  • Published, attributable work. micro1's screen leans on specifics: what you wrote, for which product, who read it, what the constraint was. Links to public docs, SDK references, or open-source contributions carry more weight than a title.
  • Domain adjacency. Developer docs, API references, cloud infrastructure, hardware, security, and regulated-industry documentation are the areas most often requested. Depth in one beats familiarity with all.
  • Rubric discipline. An AI interviewer will present two drafts and ask which is better. It's measuring whether your reasons are inspectable and consistent, not whether your taste matches theirs.
  • Written throughput. Some screens include a timed writing or editing exercise. Expect to work without your usual style guide open.

Logistics

Remote and asynchronous. Most engagements run as batches of tasks with a turnaround window rather than fixed hours, commonly 10–25 hours a week, with some projects offering more during ramp periods. Work is done in the client's annotation or review tooling; you'll need a reliable machine, a stable connection, and a quiet setup for the recorded AI screen. Rates in the $90–140 band are what contributors have reported; the actual offer depends on the project, the domain, and how the calibration round goes.