Every outbound agent payment has to clear six separate checks before it leaves the wallet: allowlist, daily cap, per-recipient cap, price window, trust tier, and anomaly. If even one check fails, the payment doesn't get signed. Each check leaves its own fingerprint in the receipt.
Is the recipient on the agent's approved list? If we don't know them, the payment never gets signed.
Has the agent already hit its daily spending limit? If so, this payment stops here.
Has this one recipient already gotten their max for today? This stops someone from draining a wallet through repeat small payments.
Is the price close to what this resource has actually cost in the last few minutes? This blocks someone from feeding the agent a rigged price.
Does the recipient's HiveTrust score meet the bar for this kind of payment?
Does this payment look like the agent's normal behavior, or does it stand out as unusual?
A single check fails in the real world. People talk their way past allowlists. Spending caps get split across several recipients to dodge the limit. Price feeds get manipulated. Trust scores go stale. SHOD runs six checks that don't depend on each other, so getting past one isn't enough. Getting past all six at once is hard enough that it makes agent payments worth insuring.
Every payment receipt records which checks passed and by how much room. Auditors can redo that math themselves using the same on-chain data.
Referenced in the HiveAudit section of hivetrust.json. The receipts come from the srotzin/hive-passport SHOD module.