Benchmark RadarAI BENCHMARK PROFILE
Reconstruction
Reconstruction is a blind benchmark for recovering research ideas from pre-publication bibliographies, using a judge model to match hypotheses.
- Released
- 2026-08-17
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
It tests idea recovery and anti-leakage protocols, but focuses on a specific research question rather than a general comparison.
Motivation
Can a language model recover the true research idea of a published paper when given only that paper's pre-publication bibliography?
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.