Proof-Carrying Decisions

A FERZ-defined primitive in the runtime authorization framework, introduced in On the Impossibility of Observability-Based Authorization (Meyman, 2026) and developed across the FERZ corpus.

What a PCD records

A PCD is produced at the moment an AI action is authorized, denied, or abstained from. It records the verdict (ALLOW, DENY, or ABSTAIN), a reference to the policy evaluated, a binding to the inputs considered, and a cryptographic signature from the runtime authorization gate that issued it. The record is signed at production. It is not a log entry written after execution. It is not a trace reconstructed from observation. It is the decision itself, rendered as a portable artifact.

Independent verification

Because the artifact is cryptographically signed and contains both the policy reference and the input binding, verification does not require access to the executing environment. A regulator, auditor, counterparty, or downstream system can verify a PCD using only the artifact and the public verification key. The FERZ corpus refers to this property as structural independence: the authorization decision is severable from the system that produced it.

Same shape across three verdicts

Every PCD carries the same record structure regardless of verdict. The canonical verdict space contains three verdicts: ALLOW, DENY, and ABSTAIN. A DENY is as verifiable as an ALLOW. An ABSTAIN verdict, which is fail-closed and blocks execution pending authorized human override, produces a PCD with the same cryptographic guarantees as any other verdict. The verdict space has three values. The artifact space has one shape.

How it differs from logs, traces, and audit records

PCDs are not audit logs, observability traces, or forensic records. Logs and traces are produced after the fact and require trust in the recording system. A PCD is produced at the authorization moment and is verifiable without trust in the executing environment. PCDs also differ from identity and access management artifacts: IAM tokens authorize principals to access resources; a PCD records the authorization of an AI action against encoded policy. Observability describes what happened. Logs record events. A PCD is the decision artifact itself.

Where it sits in the framework

PCDs are the artifact output of execution-time authorization at the runtime authorization boundary. Where the boundary specifies the structural location of authorization and execution-time authorization specifies the temporal moment, the PCD specifies the form in which each decision is recorded and made independently verifiable. Together with the verdict space and the principle of non-bypassable governance, Proof-Carrying Decisions complete the artifact layer of the runtime authorization framework.

References

1. Meyman, E. (2026). On the Impossibility of Observability-Based Authorization, v1.3.2. FERZ, Inc. https://doi.org/10.5281/zenodo.19647542
2. Meyman, E. (2026). Execution-Time Authorization for AI Agents: A Formal Framework for Deterministic Governance Boundaries. FERZ, Inc. https://doi.org/10.5281/zenodo.18764562
3. Meyman, E. (2026). Authorization Artifact Test, v1.0. FERZ, Inc. https://doi.org/10.5281/zenodo.20013582

The full FERZ corpus is available at /research/papers, on Zenodo, and on SSRN.

Related concepts

• Execution-time authorizationthe temporal property that produces PCDs
• Runtime authorization boundarythe structural location where PCDs are emitted
• ABSTAIN verdictthe fail-closed verdict that produces a PCD blocking execution
• Fail-closed designthe property an ABSTAIN PCD enforces
• Deterministic authorizationthe basis on which PCDs are reproducible
• Observability is not authorizationthe doctrinal position about why PCDs exist

Frequently Asked Questions

@misc{ferz2026pcd,
  author    = {{FERZ, Inc.}},
  title     = {Proof-Carrying Decisions: Definition and Architecture},
  year      = {2026},
  month     = {May},
  publisher = {FERZ, Inc.},
  url       = {https://ferz.ai/governance/concepts/proof-carrying-decisions}
}