Scientific Computing Expert: Physical Sciences ($300/Approved Task)

What you'll do

You convert physical-science work into benchmark environments for Terminal Bench Science. A good task has realistic scientific inputs (molecular structures, simulation parameters, experimental data or model configurations), an expert solution you wrote yourself, and automated tests that decide whether an AI agent got it right.

What the ad asks you to do:

• Design multi-step tasks from realistic physics, chemistry, materials or astronomy workflows, in self-contained environments with pinned dependencies
• Implement solutions in Python, Bash, C/C++, Julia or domain tools, covering simulations, numerical modelling, data fitting, optimisation, spectroscopy, molecular analysis or visualisation
• Write tests for numerical accuracy, physical consistency, convergence and output structure, with sensible tolerances, units and boundary conditions
• Confirm tasks run without runtime downloads and debug dependency, precision, solver and file-format problems

The "no runtime downloads" rule is a practical point: the task has to work offline, so bundling data and tools correctly is part of the craft.

Who fits

• PhD, postdoctoral or equivalent advanced experience in a relevant physical-science discipline
• Strong programming in Python, C/C++, Julia, Bash or similar
• Linux and terminal experience
• Numerical methods, scientific modelling, simulation or quantitative data analysis
• Able to build and validate computational workflows independently

What it pays

$300 per approved task, as stated. Siblings from Turing (Earth Sciences, Engineering, Mathematics, Life Science) carry the same figure. Pay depends on tasks being approved; no quota is given.

Worth knowing

Good:

• Published rate
• Broad eligibility across physics, chemistry, materials and astronomy
• Authoring work with a clear pass or fail

Less good:

• Per-task pay with review risk
• At least 6 hours a day and 40 hours a week, 4 hours of Pacific Time overlap
• 5 weeks, start expected the following week, no medical or paid leave
• Assessment steps are not in the ad; Turing's help centre describes skill tests and, for some roles, live interviews

About this listing

Posted by Turing on 25 September 2026 and read on 18 October 2026. Rate and terms are Turing's own text; the notes on offline tasks and sibling comparison are ours. See Turing.