What the work involves

You will spend your days in two adjacent modes. The first is production: simulating raw patient-level data that behaves like a real Phase 1–3 study — plausible enrollment patterns, dropouts, protocol deviations, missingness that isn't uniform — then mapping it into SDTM domains, deriving ADaM analysis datasets, and generating the tables, listings, and figures. The standard is traceability: a number in a TLF must be walkable back through ADaM derivations to the SDTM record to the source observation. You are working from an existing SAP, not writing one; the judgment being asked of you is interpretation and implementation fidelity, not study design.

The second mode is task authorship. Once you can do the pipeline, you decompose it into tasks an AI agent might attempt — full end-to-end runs and narrow component steps like a single ADaM derivation or a specific summary table. For each you write the prompt, the golden output, and a rubric that distinguishes a correct result from one that looks correct. That rubric work is the hard part: deciding which deviations are cosmetic, which break traceability, and how partial credit should behave when an agent gets the population flag right but the imputation wrong.

What the platform screens for

Mercor's screen is AI-led and follow-up heavy. Expect to be pushed on specific domains and variables rather than asked whether you "know CDISC" — naming ADSL population flags, explaining how you handled a particular SUPPQUAL or RELREC situation, walking a define.xml decision. Claims about pre-market Phase 1–3 experience get tested with detail questions, so bring studies you can actually describe. Evaluation judgment is probed separately: how you would grade an imperfect agent output, and whether you can articulate a failure taxonomy rather than a pass/fail instinct.

Logistics

  • Remote, US or Canada only.
  • 20–25 hours per week minimum; 30+ preferred and generally means steadier task flow.
  • Largely asynchronous, with some overlap expected for calibration discussions with other statisticians on the project.
  • Pay observed in the $120–170/hr range, varying with experience and the mix of production versus authorship work. Not guaranteed.