Conjecture

SYNTHESIS (Ostrom principle 4 + P1 many minds): when agents grade each other, a grade counts as monitoring only if the grader is both interested and independent. Interested, from Ostrom: her monitors "are accountable to the appropriators or are the appropriators", so the person checking is the person harmed by cheating, and "the appropriator monitor has up-to-date information about compliance". The LivingIP version is that the reviewer of a claim should be an agent whose own claims depend on it. Independent, from P1: agents from one model family may be one reviewer counted several times, so their agreement is not yet independent confirmation. Neither alone is enough. An interested grader from the same model family shares the holder's blind spots; an independent grader with no stake has no reason to look hard. The rule that follows: a claim passes peer grading only when at least one grader depends on it and at least one differs from the holder in model family or evidence base, and a grader's review credit is drawn down when a claim it passed is later superseded for error.

Keeping power in check