What the work actually is

You take a piece of computational work you have genuinely done — a simulation you set up, a numerical scheme you debugged, an analysis pipeline you built over a dataset — and rebuild it as a standalone evaluation task. That means writing a problem statement a competent researcher in your subfield could act on without asking you clarifying questions, supplying whatever inputs the task needs, producing a reference solution you can defend, and specifying the metrics by which a submitted answer should be scored. Afterquery asks authors to define the grading dimensions but not to publish the exact passing thresholds, so the task can't be gamed by a model that reverse-engineers the rubric.

Approval is not automatic. Tasks go through structured review, and most authors iterate at least once — usually on underspecification (the prompt admits three defensible interpretations), on solutions that turn out to be reachable by pattern-matching rather than reasoning, or on metrics that can't be applied consistently by a second reader. The stated economics are $300 per approved task at roughly four hours each, which maps to the observed $70–80/hr band only if your revision cycles stay short. Volume is uncapped.

What the screen is looking for

  • Provenance. Tasks must come from work you did, not from textbooks, benchmarks, or published problem sets. Expect to be asked to describe a specific project in enough detail that the depth is obvious.
  • Specification discipline. Can you state a problem so that two experts would agree on what a correct answer looks like, and on how wrong answers fail?
  • Difficulty calibration. You need a working sense of what current models already handle well in your area, so your task lands above that line rather than below it.
  • Metric literacy. Numerical tolerance, convergence criteria, statistical significance, reproducibility across seeds — you should be able to say why a given tolerance is the right one.

Logistics

Fully remote, fully asynchronous, no fixed weekly hours and no minimum. You choose when to author and how many tasks to take. Optional peer review and calibration work is available in your domain if you want it. The practical constraints are that you need enough hands-on research to keep sourcing original problems, and enough self-management to carry a task through review cycles without being chased.