What the work involves

You are matched to one of twelve specific research areas, then assigned a role within it: creating research-level problems, solving them, reviewing completed work, or auditing it. The assignment is agreed with you after matching, and it depends on where your publication record sits. Whatever the role, the deliverable is the same in kind — a worked argument, not an answer key. Derivations need to be reproducible by another specialist without talking to you, which in practice means LaTeX write-ups, SymPy or numerical checks where the algebra is heavy, and explicit statements of the assumptions and limits under which a result holds.

The problems are meant to separate genuine research reasoning from textbook recall, so a good problem is one where a frontier model can plausibly produce something that looks right and is wrong. Auditing work means finding that failure mode — the Haar average that silently assumes the unitary group when the paper specified O(N), the QFI bound quoted without the estimator that saturates it, the channel argument that only holds in the degradable regime.

What the platform screens for

The decisive filter is narrow topical match. Mercor states plainly that the most common reason strong physicists are declined is having published adjacent to the phenomenon rather than on it: general command of the information-theoretic toolkit does not substitute for papers on, say, superactivation of quantum capacity or simplex embedding for noncontextuality. You will be asked for three to five representative papers with arXiv IDs or DOIs, ideally from the last five years, first author strongly preferred, and every one is checked against the public record. You select up to five areas and must attach your own paper as proof for each — an unproven selection is not scored.

A PhD in quantum information, quantum optics or a closely related field is a hard requirement. Working proficiency with LaTeX, Python, SymPy and Jupyter is expected; familiarity with an agentic coding extension in VS Code is a plus, and honestly declared gaps are tolerated. English at B2 or above, including written argument.

Logistics

  • Roughly 10 hours per week, sustained over an 8 to 10 week window, starting immediately
  • Fully remote and asynchronous, no fixed hours
  • $80–110/hr as observed on this listing, set by depth of subdomain match — not guaranteed
  • Process: application form with method self-attestation, publication verification, then either direct onboarding or a short live alignment call; a 30–45 minute assessment is added only where needed

Reading [arXiv:2509.26574](https://arxiv.org/abs/2509.26574) before applying is genuinely worth the hour — it will tell you quickly whether this kind of problem design appeals to you.