How to prepare for the Mercor screening test

Mercor screens every applicant before unlocking paid work. The screening test is the single barrier most candidates fail. Preparing properly increases the acceptance rate sharply. ## What the test looks like in 2026 The screening is a short timed exercise tailored to your declared domain. For software developers it usually combines a code-reading task (rate two model outputs and justify why one is better), a written-judgment prompt, and a short live or async interview. ## Domain-specific format - Software engineers: pairwise code-quality comparisons and one open-ended debugging walkthrough. - ML engineers: model-output evaluation across reasoning, factuality, and code generation. - Domain experts (law, finance, medicine, science): scenario evaluations on rubric criteria specific to the field. ## How to prepare in a week 1. Spend two evenings on Mercor's documentation. Read the public rubrics they publish for evaluators. 2. Practise pairwise judgment on free public datasets (Anthropic HH-RLHF samples, OpenAI evals, public model leaderboard outputs). Form a written opinion on each before checking the consensus. 3. Write your justifications as you would for paid work: three sentences minimum, specific to the artefact, no generic praise. ## What to write in the open-ended sections Specificity over volume. Reviewers reward sharp, falsifiable claims tied to the specific output. Vague reasoning ("the second response is more comprehensive") is downrated. Concrete reasoning ("the second response correctly identifies the off-by-one error on line 14, while the first misses it") passes. ## What to avoid Do not use an LLM to draft the written sections. Mercor's screeners explicitly check for LLM-generated text patterns and reject candidates who use them. The screen is for your judgment, not a model's. ## Re-applying after rejection If you fail, Mercor allows reapplication after roughly six months. Use the gap to publish technical writing publicly (blog posts, GitHub READMEs) that demonstrates the judgment they screened for. Profile depth between attempts is the strongest re-application signal.