What the work involves

You will write and verify research-level physics problems and their full solutions — the kind of material a frontier model currently fails on. Typical items require deriving field equations in the first-order Palatini formalism using tetrads and spin connections, computing Pontryagin density contributions in Chern-Simons-modified gravity, setting up Friedmann ODEs with torsion, and integrating coupled nonlinear systems numerically with consistent Planck-unit normalization. Deliverables are expected to be self-contained: statement, derivation, numerical result where applicable, and a clear account of assumptions and conventions (signature, index placement, units).

Roles are split by function. Solvers produce the analytical and numerical work. Auditors independently reproduce it and flag sign errors, convention mismatches, unstated assumptions, and problems that are ambiguous or under-determined. Adjudicators resolve disagreements between the first two and issue a binding call with written reasoning. Which track you're offered depends on your subfield depth and how you perform on the screen.

What the platform screens for

  • Verifiable depth. micro1's screen is AI-led and conversational, with follow-ups that push past a first answer. Expect to be asked to reason aloud through a derivation rather than name-drop a formalism.
  • Convention discipline. A large share of benchmark defects are convention errors, not physics errors. Screeners probe whether you state and hold conventions consistently.
  • Evaluation judgment. Can you tell a wrong answer from a differently-conventioned right answer? Can you say when a problem is ill-posed rather than forcing a solution?
  • Provenance. Publications, thesis topic, and specific systems you've worked with are checked against what you claim in conversation.

Logistics

Fully remote, contractor engagement, asynchronous. Contributors commonly report 10–25 hours per week with flexible scheduling, though project phases are bursty and volume can drop between customer batches. Pay in the $80–160/hr range has been observed for this listing; actual offers vary by track (adjudication typically sits higher than solving) and by subfield scarcity, and nothing here is guaranteed. No prior AI or ML experience is required — LaTeX fluency and comfort with a numerical stack (Python/SciPy, Mathematica) matter far more.