What problem does it solve? Choosing between composition topologies (e.g. deep vs wide research orchestration) over the same atomic skills is guesswork without measured data. This Skill runs each method over a shared task set and seeds, judges outputs against a contract rubric, measures cost, and produces a Pareto frontier with a statistically grounded winner. ## Core Features & Use Cases - Multi-method benchmarking: Runs each composition method end-to-end across tasks and seeds via the run-composition skill, recording agent and token costs per run. - Rubric-based LLM judging: Dispatches a judge agent per output with the contract rubric, computing the overall quality score in the harness as the unweighted mean of accuracy, coverage, citations, and freshness subscores. - Pareto frontier analysis: Aggregates quality and cost per method, then uses the compose-eval CLI to compute the Pareto set and winner, writing composition-profile.json. - Use Case: Compare a wide-fanout research orchestration against a lean-linear pipeline on the same research questions to find which delivers the best quality per token budget. ## Quick Start Compare the research-orchestration and research-lean-linear composition methods on the research contract and show me the quality-cost frontier.
