What the work involves

CritPt (see [arXiv:2509.26574](https://arxiv.org/abs/2509.26574)) tests whether frontier models can do genuine physics research reasoning rather than textbook problem solving. Depending on where your publication record sits, you may be authoring new research-level problems in stochastic autocatalytic growth, producing full solutions, reviewing another contributor's work, or auditing a completed item. The assignment is agreed with you after matching, not assigned blind.

Every deliverable carries written reasoning. A problem must be unambiguous and have a defensible reference solution; a solution must be reconstructible by another specialist in reaction-network stochastics without emailing you. Expect to work in LaTeX, with Python, SymPy and Jupyter for symbolic checks — generating-function manipulations, circulant spectra, moment closure sanity checks. Familiarity with an agentic coding extension in VS Code helps but is not required if you say so up front.

What the platform screens for

The gate is narrow and it is publication-based. The panel covers one phenomenon — stochastic autocatalytic growth: the stochastic Hinshelwood cycle, chemical master equation processes, circulant matrix spectra, transient oscillations toward balanced exponential growth, moment hierarchies, generating-function methods. Having the methods in your toolkit is not sufficient; the most common decline is a strong researcher who has never published on the phenomenon itself. You will be asked for three to five papers with arXiv IDs or DOIs, your author position, and which methods each one demonstrates. Authorship is checked against the public record, and a selection without proof is not scored.

A PhD in biophysics, statistical physics, applied mathematics or a closely adjacent field is a hard requirement. Postdocs, research scientists and junior faculty are the strongest fit; senior PhD students with a solid first-author record are welcome. English at B2 or above, including in writing.

Logistics

  • Remote, fully asynchronous, no fixed hours
  • 10 hours per week sustained across an 8–10 week window, starting immediately
  • $80–110/hr as observed on the platform, set by depth of subdomain match — not guaranteed
  • Process: form with method self-attestation and proof paper → publication verification → direct onboarding or a short live alignment call; a 30–45 minute assessment is added only in some cases

If you also work on long-range dispersal, heavy-tailed jump processes or front propagation in range expansions, flag it — a related area sits in the statistical physics panel and uses the same machinery.