How we write our questions
What we do, what we do not do, and why you still need to read each platform's live guidelines.
The short version
Every question on AnnotationPrep is original. Scenarios are crafted to mirror the failure modes that trip up applicants—subtle hallucinations, missed negative constraints, flawed chain-of-thought math, buggy edge cases, and non-atomic rubric criteria—without copying any proprietary platform test.
How a question is made
- Original scenarios only. Every prompt, Model A/B comparison, code snippet, and rubric item is written fresh.
- Clear grading principle. Each answer cites the exact evaluation rule, debugging principle, or rating standard it rests on.
- Format and uniqueness checks. Every item is validated for a single unambiguous best answer and distinct distractors.
- Technical and math verification. Coding edge cases, complexity bounds, and mathematical proofs are verified independently.
- Independent second check. Answers and rationales are reviewed against the cited evaluation principle.
- Community reports. Reports sent with “Report this question” are reviewed and acted on quickly. See Corrections.
What we do not do
- We do not copy, scrape, or leak real proprietary assessment questions from any platform.
- We do not publish pass rates or guarantee acceptance onto any platform.
- We do not encourage using AI tools during live platform assessments—practising here builds your own speed and judgment so you can pass legitimately.
Why platform guidelines still come first
Each AI training platform and each individual project has its own specific style guide, priority hierarchy (e.g., Truthfulness vs. Instruction Following), and formatting rules.
Use AnnotationPrep to train your eye for errors and build speed, then read the live instructions on the platform before clicking Submit.
If you find a mistake
Use “Report this question” under any question, or email support@annotationprep.com. Corrections explains what happens next.