Computational Statistics and Applied Mathematics Expert (R, Python, and Matlab/Scilab)

$70–90/hrrate as listed

Design original graduate-level computational problems that test whether frontier AI models can use specialized statistical and mathematical software to do research-level work. Each problem is built, tested against state-of-the-art models, and iterated until it lands at the target difficulty.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

The work

This is task authorship for a large-scale scientific benchmark, not annotation. You pick a domain you genuinely know — Bayesian inference in Stan, IRT in mirt or TAM, spatial point processes in spatstat, state-space models in KFAS or MARSS, survival analysis in flexsurv, ODE inference in pomp or deSolve, high-precision linear algebra in Rmpfr, or the Matlab/Scilab equivalents — and build problems that require a model to actually run the software correctly rather than pattern-match a plausible answer.

Two broad problem types recur. The first is fully specified and reproducible: a defined setup with a single correct numerical answer that can only be reached through a multi-step workflow. The second is interactive and harder: the model must plan a sequence of queries or simulated experiments to recover information that is not directly observable, reason from partial results, and narrow the hypothesis space efficiently. Both types require you to write the setup, an oracle function, and a validator in Python, then run the problem against current frontier models and revise until the difficulty is right — too easy and it gets cut, too ambiguous and it gets cut.

What the screen looks for

Verifiable depth in a specific package. Expect follow-ups on convergence diagnostics, identifiability, parameterization choices, numerical stability, and where your tool of choice quietly gives wrong answers.
Evidence, not claims. Publications, open-source contributions, or professional work that show you have used these packages on real research problems.
Problem-design instinct. Can you distinguish a hard problem from a merely tedious one, and can you specify a task tightly enough that a validator can judge it?
Python fluency and terminal comfort. You will write oracles and validators and work inside Linux-based remote compute sandboxes.

Logistics

Fully remote and asynchronous, with roughly 15–20 hours per week expected. Credentials required are MS or PhD in statistics, applied mathematics, or a closely related quantitative field — PhD preferred, or MS with 10+ years of relevant experience. The observed pay band is $70–90/hr; rates on Mercor vary by domain, assessed depth, and project stage, and are set at offer rather than guaranteed by the listing.

What they look for

Graduate degree (MS or PhD) in statistics, applied mathematics, or a closely related quantitative field
Deep hands-on expertise with at least one specialized statistical/mathematical package (e.g. Stan, mirt, lavaan, spatstat, KFAS, deSolve, mgcv, flexsurv, PyMC, statsmodels, Matlab/Scilab toolboxes)
Strong Python for writing problem setups, oracle functions, and solution validators
Demonstrated research or professional output using those tools (publications, open-source contributions, or applied work)
Comfort in Linux/terminal environments and remote compute sandboxes
Problem- or exam-design experience, or prior benchmark/evaluation work (nice to have)

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Do you hold a Master's degree or higher in statistics, applied mathematics, computer science, physics, engineering, economics, or a closely related quantitative field? (Or, if your highest degree is a Bachelor's, do you have 10+ years of full-time research or professional experience in computational statistics or applied mathematics?)
Graduate degree in quantitative field · required
Have you written and run code using at least one of the specialized statistical, mathematical, or scientific packages listed in the posting (such as rstan, lavaan, spatstat, MARSS, PyMC, or similar domain-specific tools) as part of published research, open-source contribution, or professional work?
Hands-on experience with a specialized statistical or mathematical package · required
Can you write Python code to set up computational problems, build validation logic, and test solutions—i.e., do you have production-grade Python skills beyond data analysis notebooks?
Strong Python proficiency · required
Are you comfortable working in a Linux command-line environment and familiar with remote compute sandboxes or containerized workflows (Docker, singularity, or similar)?
Linux and terminal comfort · preferred
Can you commit at least 15–20 hours per week to this project over the contract period?
Availability for 15–20 hours per week · required

0 of 5 answered

Included with this listing

Interview prep — 11 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether you have genuine hands-on research experience with domain-specific tools, not just theoretical familiarity, and whether you understand where these packages excel and where they fail.
A strong answer: Name a real project with a specific scientific question, describe the package you chose and why, explain a non-obvious decision or edge case you hit, and reflect on what the package couldn't do or where you had to work around its design.
Where people lose it: Listing package names without a concrete example, or describing generic statistical tasks (fitting a linear model, running a t-test) rather than specialized, research-grade workflows that demand deep package knowledge.