An argument prints as a claim-and-evidence graph with per-claim link counts. A reader panel scores versions by what readers actually restate. And the same content in two structures wins different tasks on different printed tables. Commit a prediction before every run.
Choose a mission to open its prediction gate.