What the work involves
You will spend most of your time producing and stress-testing algorithmic problems that models find hard. In practice that means authoring original problems with clean statements and tight constraints, writing a reference solution plus at least one deliberately wrong-but-plausible solution, generating adversarial test cases that separate an intended O(n log n) approach from an O(n²) one, and documenting why a model's submission fails — TLE on a specific input class, an off-by-one in a binary search predicate, a wrong greedy exchange argument. Some tasks flip the direction: you are handed model-generated code and asked to judge correctness, find the counterexample, and write a short rationale a non-author can verify.
What the platform screens for
AfterQuery's screen is credential-anchored and then depth-tested. Expect to state your handle and peak rating, ICPC placement, or USACO division — figures that can be checked against public profiles. After that, the questions get technical fast: describe a specific problem you solved and the key insight, explain how you would break a given solution, distinguish an amortised bound from a worst-case one. The screen is looking for people who can articulate reasoning, not just produce accepted submissions. Written clarity matters here more than it does in a contest; a correct verdict with an unreadable justification is not a usable data point.
Logistics
- Fully remote, asynchronous, no fixed hours
- 10–20 hours per week, sustained over the project rather than burst
- Observed pay in the $30–60/hr range, varying with rating tier and task type; not guaranteed
- Work is delivered through the platform's task queue with review cycles — expect revision requests on early submissions while calibrating to the rubric
- Comfort with English technical writing is assumed, as problem statements and failure analyses are the deliverable