CJ Receipts · Deep Research Benchmark

Cost per verified claim.

Public benchmarks reward reaching the right answer. None of them grades whether a research report's claims are true and its citations real. CJ Receipts audits imported deep-research reports claim by claim, with evidence-only judges from a different vendor than the author.

Small-n by design. Every score shows its 95% bootstrap interval, the number of runs, and whether a hard cap fired. Scores marked provisional still have disputed claims awaiting a human label.

No graded runs published yet. The first set, the Gemini 4 Argon launch (2026-09-30), is being graded: a Cartographer leg and an Adversary leg, each audited by judges from other vendors. Nothing here is estimated or placeholder. Read the methodology to see exactly how it will be scored.

Generated 2026-10-01T08:31:59+00:00 · 0 graded runs