Most AI work leaves no trace. A receipt rail records the material steps of an AI-assisted workflow so the work can be inspected, audited, and disputed later.
Without receipts, post-incident review is forensic guesswork.
Email threads and Slack messages aren't a chain of custody.
A clean run looks the same as a hidden retry.
Regulated and enterprise buyers expect to see the chain.
| Field | What it captures |
|---|---|
| Actor identity | The human or service account that initiated the run. |
| Agent identity | The agent or operator that executed the run (e.g., Iris Meridian). |
| System touched | The system of record changed or read (CMS, CRM, DAM, etc.). |
| Policy pack / version | The signed bundle of rules bound to the run. |
| Model / runtime | Which model and which runtime location handled inference. |
| Execution location | Local, hybrid, or cloud — proven, not asserted. |
| Input hash | Cryptographic hash of the input artifacts. |
| Output hash | Cryptographic hash of the produced artifacts. |
| Approval state | Who approved, when, and at which step. |
| Action allowed or blocked | Each material action recorded with verdict. |
| Reason code | Why an action was allowed or blocked. |
| Outcome grade | Brand / locale / safety / provenance grade. |
| Receipt hash | Hash that anchors the receipt for later inspection. |
Receipts are addressable, indexable, and queryable by workflow, system, policy, and approver.
Each receipt's hash anchors the chain. Tampering breaks the seal.
Receipts know which policy pack version applied at runtime.
Receipts record who approved which step and when.
Receipts carry the outcome grade dimensions.
Receipts reflect revocation state at the time of the action.
A receipt is a sealed, addressable artifact. It survives review. Logs decay; receipts hold up under audit, dispute, and incident reconstruction.
The Receipt Rail Sprint binds one high-risk AI workflow end-to-end and delivers a sealed packet.
