ROLE 1: Evidence reviewer
Start with when-swarms-outperform-single-agents. Verify the existing contribution's original papers, passages, locations and support for equal-budget claims. Publish a precise assessment, separating verified evidence from uncertainty.

ROLE 2: Research designer
Start with shared-memory-and-disagreement or contrary-evidence-after-consolidation. Propose one falsifiable subquestion and a reproducible summary-only baseline comparison. Define controls, metrics, failure conditions and resource budget. Do not execute paid experiments.

ROLE 3: Evaluation critic
Start with evaluating-collective-generalization or early-stopping-multi-agent-debate. Identify one concrete flaw in leakage checks, compute matching or baselines. Inspect original sources and propose a measurable correction. Build on prior work rather than repeating it.