What the work actually involves

This is a calibration study, not ongoing annotation work. You sign an NDA, receive a packet of source files — the kind of raw material a seller actually works from, such as call transcripts, CRM exports, discovery notes, pricing sheets or an RFP — and produce one realistic deliverable yourself. That first-hand pass matters: it establishes what a competent professional would do before you see any model attempt. You then review a set of model-generated outputs for the same task, rank them, score them against the study's rubric, and write rationales that point to specific evidence in the source files rather than to general impressions.

The final component is structured feedback on the study design itself: whether the task was representative of real sales work, whether the rubric captured what separates good output from bad, and where the materials were ambiguous. Mercor treats this feedback as part of the deliverable, not an afterthought.

What the screen is looking for

  • Depth of practice, not job titles. Five-plus years hands-on and currently or recently selling. Expect follow-ups on your actual motion — deal sizes, cycle length, whether you ran discovery yourself or inherited qualified pipeline.
  • Evidence-grounded writing. Rationales that cite the transcript line or the pricing constraint that makes one output wrong. Vague preference language scores poorly.
  • Discrimination between near-identical outputs. Much of the value comes from separating two outputs that both read fluently but differ on one substantive point.
  • Honest availability. The window is fixed. Partial completion is worth little to a calibration study.

Logistics

Fully remote and asynchronous within the September 14–18, 2026 window; you schedule the 13 hours yourself. No LLMs or AI assistants may be used at any point during the study — the whole purpose is to capture unassisted human judgment, and violations invalidate the data. Anyone who has worked on Project Alchemy in any capacity, or accessed its source-world data, is ineligible. Pay is observed at $90/hr for this listing; rates on Mercor vary by study and are not guaranteed.