These are four moments where a portable, offline-verifiable credential matters more than a dashboard, a screenshot, or a self-reported log stream.
A Verifiable Process Execution Credential is a cryptographic record proving that defined governance checks ran on specific data — verifiable offline, without asking your dashboard for permission. Ed25519-signed, RFC 3161 timestamped, content-blind by construction.
These are four real moments where having a VPEC instead of a log makes the difference. If one of them sounds familiar, that's where to start.
Under the current EU implementation timeline, high-risk AI obligations for Annex III systems and transparency rules apply from August 2, 2026. An auditor asks for evidence that your AI governance checks ran on a specific deployment. You show them dashboards, log files, screenshots. The auditor says: how do I know these checks actually ran on this data, on this date, and produced this result? You don't have an answer.
The artifact gives them something to test: signature, timestamp, committed inputs and outputs, declared checks, proof floor, and gaps. They verify the artifact instead of relying on your dashboard narrative.
Your reinsurer needs to verify that claims adjudication ran per treaty terms. The standard request is the claim file. But the claim file contains policyholder PII. You can't send it. They can't verify without it. This is the structural proof problem that keeps reappearing in insurance workflows.
A VPEC from Guidewire ClaimCenter proves adjudication ran per treaty terms — without the claim file ever leaving your environment. The reinsurer verifies the credential. The chain is intact. The stalemate is over.
Your OPA policies may already guard every request. Today that gives you authorization and logs: useful, but not a portable proof artifact. Add Primust at the evaluation boundary and each new decision emits a VPEC bound to the policy version, input commitment, output, and timestamp.
OPA evaluations are deterministic: same input, same policy, same output. That determinism is why an instrumented eval() can reach Mathematical proof. One Go package, no policy rewrite, no platform replacement.
An insurer is writing an AI liability policy. They ask for evidence that your governance controls are effective — not that they're configured, but that they actually fired and blocked something. Self-reported dashboards aren't auditable. Log files can be manipulated. There's no CVSSv3 for AI governance quality. They can't price the risk.
VPECs can become that measurement layer. blocked_count > 0 shows controls fired. provable_surface over 90 days gives underwriters cleaner runtime data than screenshots or self-reported dashboards.
blocked_count > 0 shows controls fired. provable_surface over 90 days is machine-readable runtime data. Not self-reported. Not a dashboard. A cryptographic record.pip install primust && primust init. Sandbox key, free and unlimited. No re-instrumentation when you go to production. The verifier is Apache-2.0 and runs without a live Primust API.