Loading…
Policy pack
Affiliate Rewards
Governance for affiliate/referral funnel agents that issue reward and commission payouts.
- Key
- affiliate_rewards
- Version
- affiliate_rewards-v1
- Rules
- 4
- Domain
- AFFILIATE_REWARDS
Rules, in the order the evaluator tries them
Highest priority first. The first rule whose conditions all hold decides; with none, the decision is a REVIEW.
Reject structurally invalid negative payout amounts
priority 200reject_negative_amount · AFFILIATE_REWARDS
When- amount < 0
ThenAUTO_REJECT → REJECT · urgent · SLA 60 min
Negative reward amounts are invalid and must never be paid.
Review elevated-risk reward payouts
priority 150review_high_risk_payout · AFFILIATE_REWARDS
When- risk_score ≥ 0.6
ThenREQUIRE_REVIEW → REVIEW · elevated · SLA 240 min
Elevated fraud/risk signals hold the payout for accountable review.
Review large reward payouts
priority 140review_large_payout · AFFILIATE_REWARDS
When- amount > 100
ThenREQUIRE_REVIEW → REVIEW · elevated · SLA 240 min
Large payouts (in the funnel's amount unit) get a manual accountability check.
Approve standard referral reward payouts
priority 20approve_standard_payout · AFFILIATE_REWARDS
When- decision_type = "REFERRAL_REWARD_PAYOUT"
ThenAUTO_APPROVE → APPROVE · routine · SLA 30 min
Standard low-value, low-risk referral rewards are credited automatically.
Evaluate under it
Locally, with @decionis/mcp: the payload becomes the request's context, and its action the decision type. On the hosted protocol, once the bundle is stored for your organization, name vertical_pack, policy_version and require_exact_policy_version: true.
decionis_evaluate arguments
{ "policy_pack": "affiliate_rewards", "payload": { "action": "<decision type>" } }In a hosted evaluate-decision request
"vertical_pack": "affiliate_rewards",
"policy_version": "affiliate_rewards-v1",
"require_exact_policy_version": trueWorkflow keys that resolve to this pack: reward_payout, referral_reward, affiliate_commission.
What its rules read
amount, decision_type, risk_score are fields of the request itself, not of its context. A value put only in the context is discarded, and the evaluator holds the decision as an ESCALATE rather than decide without it. @decionis/mcp puts the whole payload in the context, so a payload that carries amount, risk_score is held.
- amountrequest
- decision_typerequest
- risk_scorerequest
Signals it maps
- payout_amountfunnel
- fraud_risk_scorerisk