What the work actually involves

You start from your own working life. A three-statement model with circular interest, a board deck that has to survive a CFO's questions, a 40-page RFP response with tracked changes from four reviewers — these become the prompts. You write the scenario with enough business context that a model can't succeed by pattern-matching a template, then run it against language models, sometimes in multi-turn conversation where you push back the way a demanding stakeholder would. When two or more model outputs come back, you rank them and write the reasoning: which formula structure would break on next quarter's data, which slide buries the ask, which Word output looks polished but misstates the covenant.

The deliverable is rarely the document. It's the assessment — a written record of what the model got right, where it failed, and why that failure would matter to someone whose name is on the deliverable. Expect to spend as much time articulating rationale as producing artifacts.

What the screen is looking for

  • Verifiable functional depth. Three-plus years in a real business function, and specifics: what you modeled, who consumed the deck, what the reporting cadence was. Vague "I'm advanced in Excel" claims collapse under one follow-up.
  • Task-design instinct. Can you invent a task that is genuinely hard for a competent model but unambiguous enough to grade? That's a different skill from doing the work.
  • Evaluation discipline. Preference for one output isn't an evaluation. The screen probes whether you can name the criterion, apply it consistently, and separate cosmetic polish from substantive correctness.
  • Writing that holds up. Feedback other people act on, not adjectives.

Logistics

Contractor engagement, fully remote, no fixed hours. Coordination with project coordinators happens asynchronously — you'll get task specs and calibration notes, and turnaround expectations are measured in days rather than hours. Volume typically fluctuates with project phase, so treat it as variable part-time work rather than a stable weekly block. Pay in the $35–50/hr range has been observed for this role; actual rate depends on the specialization and calibration outcome, and is set by the platform.