Why agent work needs receipts
When an autonomous agent pays for work, the interesting question is rarely only whether money moved. The buyer needs to know what was promised, which output was delivered, who judged it and what settlement evidence exists.
A receipt layer turns that lifecycle into machine-readable records. That makes the workflow inspectable by another agent, a wallet, a registry, a verifier, a dashboard or a later audit.
The three-record model
The Accord Agreement records terms before work starts. The Verification Receipt records whether the work was accepted, rejected or partially accepted. The Settlement Receipt records the economic closeout on the chosen rail.
Keeping those records separate is deliberate. Payment proof, work judgment and settlement evidence have different trust boundaries and often come from different systems.
Receipts are policy inputs
A buyer wallet can use an agreement hash before granting payment authority. A provider registry can use verification receipts to show what kind of work has passed. A monitoring system can compare settled work against accepted work and flag drift.
The point is not to make every workflow trustless. The point is to make the trust claims explicit enough that software can reason about them.
What makes a receipt useful
A useful receipt binds to the correct agreement, names the verifier or rail, includes stable output or transaction references and can be validated against a schema. If the receipt is signed, the signature should cover the canonical receipt object.
Receipts should avoid vague prose where policy needs structured facts. Human notes are helpful, but the core verdict and settlement state need to be parseable.
Current safety posture
Accord v0 is alpha / testnet-first and NOT CERTIFIED FOR MAINNET. The receipt model is ready for demos, conformance testing and testnet evidence, but real-fund production workflows remain audit-gated.
That conservative posture is part of the design. A receipt layer becomes more valuable when it refuses to blur compatibility, testnet success and production safety.
FAQ
Is Accord a payment rail?
No. Accord can bind to rails, but its core role is the work agreement and receipt layer around payment and settlement.
Why not use one combined receipt?
Separate records keep payment proof, work verification and settlement evidence from being confused with each other.
Can another agent read Accord receipts?
Yes. The goal is machine-readable agreement and receipt objects that agents and policy engines can inspect.
Sources and references
Accord Agreement schema
Public schema for the agreement object that anchors the receipt trail.
Accord Verification Receipt schema
Public schema for the work-completion verdict record.
Accord Settlement Receipt schema
Public schema for recording rail settlement evidence.
Inspect the receipt schemas
Follow the Agreement, Verification Receipt and Settlement Receipt trail.