Automation, policy, and fraud controls

Raidz may automate campaign planning, evidence collection, scoring, verification assistance, attribution, and settlement preparation. Automation is bounded by policy, consent, explainability, and human review; it does not automate social engagement or bypass wallet intent.

Policy engine

The implemented baseline evaluates channel, service, jurisdiction, category, compensation, copy, policy source, and effective period. It produces ELIGIBLE, BLOCKED, or—at the rule-model level—REVIEW_FLAG, plus reason codes and policy version.

The current API preflight returns blocked when any implemented reason matches. Rule sources and effective dates must be refreshed as platforms and laws change. The engine is a control input, not legal advice.

Anti-Sybil signals

The planned integrity engine groups signals without treating any single heuristic as proof:

• Social: account-age anomalies, repeated audience clusters, suspicious growth, copied content, improbable reach, coordinated timing.
• Wallet: shared funding sources, rapid wallet rotation, repeated transaction patterns, related addresses, circular flows.
• Campaign: duplicate submissions, hash reuse, creator/project collusion, evidence outside windows, repeated low-quality outcomes.
• Marketplace: review rings, self-dealing, related-party bookings, fee farming, abnormal cancellation/refund patterns.

High-risk actions require stronger corroboration. Related-party and operator activity must be labeled, not counted as independent traction.

Fraud response

Possible outcomes are allow, request more evidence, place a non-settling review flag, reject under a known reason, suspend a surface, or escalate an incident. Models and operators must not silently confiscate reserved creator value. Any slashing/bond system remains unimplemented and requires explicit terms and contracts.

AI campaign agent

The future agent may help a project translate an objective into audience, creator mix, deliverables, budget, schedule, and measurement plan. It should:

• show assumptions and confidence;
• use historical Raidz evidence only with proper consent and provenance;
• keep project approval before publication or funding;
• run policy preflight and surface blocked/review reasons;
• never fabricate creator availability, reach, conversion, or expected returns;
• never sign, fund, deploy, or retry a transaction under ambiguous state.

AI assistance remains planned until enough real marketplace history exists to evaluate whether it improves outcomes.