The trust package has an explicit decision vocabulary: ALLOW, RATE_LIMIT, CHALLENGE, REQUIRE_REAUTH, REQUIRE_HUMAN_APPROVAL, TEMPORARY_RESTRICTION, AUTOMATION_REVOKED, REVIEW_REQUIRED, and SUSPENDED.
Its current evaluator is intentionally small and deterministic. Later waves will supply explainable signals such as action velocity, duplicate content, device/session anomalies, challenge failures, and credential revocation. Probabilistic signals alone must not permanently ban an account.
The live follow and Contact routes perform authentication, ownership/recipient checks, block checks, and same-origin browser protection. Contact requests additionally have a deterministic velocity guard: after ten requests in a rolling 24-hour window, the eleventh attempt creates a server-side restriction lasting 24 hours. The client receives an exact reset time and may dismiss the notice, but dismissal never bypasses the restriction. This is intentionally enforced in the write transaction, rather than trusting a browser-side counter.
The current guard is a targeted first control, not a substitute for the planned distributed rate limiter, trust-engine signals, or append-only audit ledger. Those broader controls remain tracked as MISSING in the implementation matrix and must close before the social graph can be called fully production-ready at scale.