Gauntlet is an open‑source pipeline that evaluates computer‑architecture papers through five expert‑persona reviewers and an adversarial synthesis stage, aiming for structured critique rather than mere summarization.

On 20 ISCA 2025 and HPCA 2026 papers, ten researchers wrote analyses, then judged each other’s work against Gauntlet. In 15 of the 20 pairings evaluators preferred the pipeline, a statistically significant advantage (p < 0.01) most pronounced in Critical Rigor and absent only in Calibration.

A 98‑paper automated ablation shows the gain comes from the multi‑agent structure: the pipeline beats a single‑persona run on 96% of papers, especially due to its synthesis pass.

All analyses, scores, and the rubric are released as a community resource, encouraging broader adoption and further study of LLM‑driven technical review.