What the work actually is

Each task starts from a pool entry: a Go repository pinned to a specific git commit, an engineering question, positive and negative rubrics, foils, and a golden solution. Your job is not to answer the question — it is to make the question hard in the right way. That means reading the subsystem until you understand how it behaves at runtime, not just what the source says: following the goroutine that actually owns the lock, tracing which interface implementation is wired up at build time, checking the PR or issue history that explains why a retry loop looks the way it does. Then you rewrite the question so a frontier coding agent trips on it while a senior engineer with repository access could still answer it fairly and unambiguously.

The second half is rubric craft. Foils are the plausible-but-wrong answers — the reading a strong model would produce from grepping one file — and your rubrics have to reject them for the right reason, not on a keyword. You then run the full Studio loop (Check, Golden, Run, Analyze) and iterate until every stage passes, document the task in a README, and submit for review. Every task passes a second engineer and a super reviewer; returned tasks come back with feedback you are expected to address.

What the screen looks for

Mercor's screening is AI-led and pushes on specifics. Expect to be asked to name Go codebases you have actually navigated and describe a concrete runtime behaviour you had to reconstruct — consensus handling in etcd, LSM compaction in BadgerDB, middleware chaining in Traefik, whatever you genuinely worked in. Vague claims about "distributed systems experience" collapse under one follow-up. The task mix skews toward architecture and system design (42%), code onboarding (26%) and root-cause analysis (20%), so depth in concurrency, module boundaries, networking, storage engines or the Kubernetes ecosystem is what carries weight. Written precision is graded directly — the questions and rubrics are the deliverable.

Logistics

  • Contract, task-based, paid at an observed $130 per approved task — approval requires clearing two review layers, so plan for iteration on early tasks.
  • Fully remote and asynchronous; expert leads sit across US, India and Nigeria time zones.
  • Work happens in Mercor Studio, time tracked with Insightful, access provisioned through Okta.
  • Onboarding gates: identity verification, background check, signed Terms of Work and CIIAA, a tax form, and a short calibration quiz before task access opens.