Product comparisons
Features and prices are verified against public product pages on the stated review date. We separate vendor claims, direct observations, and our interpretation.
Answer-quality benchmarks
A benchmark must include permitted practice questions, subject and question-type distribution, grading rules, handling of ambiguous items, repeated-run policy, and enough aggregate data to reproduce the result.
2026 benchmark preregistration
The ExamNinja AI Answer Verification Benchmark 2026 uses 60 owner-authored text questions across mathematics, biology, earth science, and civics. It covers multiple choice, multiple select, short answer, and fill in the blank. Wording and expected answers are frozen before collection, and every item is run twice without conversation carry-over.
Correctness and explanation quality use separate 0–2 rubrics. A second reviewer must adjudicate disagreements, every partial score, and a deterministic 20% sample of remaining responses. Headline results must disclose denominators, exclusions, run-to-run agreement, failures, model or product version, dates, and limitations.
The benchmark remains unpublished while results are incomplete. We will not publish a score, chart, Dataset schema, or outreach claim until the non-sensitive result file and calculation script reproduce every aggregate.
Keyword and traffic research
Third-party volumes and traffic are estimates. We label the provider, market, date, and export settings and keep them separate from first-party Search Console or analytics data.
Study with the explanation, not just the answer
Try a free ExamNinja study tool, then verify important work against your course material.
Explore free tools