What problem does it solve? Deciding whether a document, report, spreadsheet, hypothesis, or Agent Skill is trustworthy and ready to release is hard when a single reviewer or majority vote can hide errors. This Skill turns agreement and disagreement into a traceable decision record backed by evidence rather than confidence language. ## Core Features & Use Cases - Five-role independent review: Evidence, outcome, and safety-portability reviewers run in parallel, a disruptor searches for falsifiable counterexamples, and a negotiator adjudicates from evidence instead of vote counting. - Conditional review protocol: The disruptor runs only after material agreement, while material disagreement routes directly to the negotiator; a five-way mode supports exploratory comparison with non-authoritative disruptor output. - Machine-readable review records: Produces a JSON record with claim ledgers, concordance classification, release decision (approve, approve-with-limits, defer-for-evidence, reject), and follow-up tests. - Use Case: Before publishing a research memo, run the bundled Python orchestrator in dry-run mode to generate frozen role prompts, then connect an explicit local agent adapter to collect structured verdicts and a defensible release decision. ## Quick Start Ask the agent to review a specific artifact file against one decision question, for example by running the equilibrium review on 'thesis.md' with the question of whether it is supported and ready for controlled publication.
