Security and AI systems detect, classify, summarize. They do not prove enough.
PoG is an open framework for turning security events and AI-assisted decisions into traceable, sealed, verifiable evidence. Not logs. Not dashboards. Artifacts that a security leader can defend in front of an auditor, a regulator, or a court.
Four properties, enforced by construction.
A system is PoG-conformant only if every governed decision produces an artifact satisfying all four properties. Three out of four is non-conformant.
Traceability
An unbroken chain from raw event to final decision: event, incident, analysis, human gate, evidence. No implicit links. Every transformation is recorded with its actor, human or machine.
Integrity
Every artifact is sealed with a cryptographic integrity mechanism at creation. Tampering, substitution, chain corruption and unauthorized resealing are detectable failure states, defined in the threat model.
Governance
The decision path itself is governed. AI-assisted judgments pass through an explicit, recorded human gate before they become authoritative. The gate is part of the sealed artifact, not a checkbox next to it.
Verifiability
Producing artifacts is not the point. Verifying them is. Any PoG artifact can be re-verified after the fact, including outside the system that produced it, using the published verification procedure.
From signal to sealed evidence.
PoG models the life of a security decision as a chain of typed objects. Each transition has explicit preconditions and failure states defined in the specification.
A framework is judged by its documents, not its slogans.
The normative core defines what PoG is, what can break it, and how conformance is verified. All three documents are drafts, versioned, and open for critical review.
All documents are versioned and published on GitHub, where reviews are handled as public issues. The current tagged version is v1.0.0-draft. Substantive reviews are credited in the changelog unless reviewers prefer otherwise. Email alternative: spec@pog-protocol.org.
Tested, not declared.
A vendor saying "we implement PoG" means nothing. Conformance under PoG is the result of executing the test plan against an implementation.
Positive tests
The implementation produces correct artifacts under normal conditions: valid chains, valid seals, valid gates, re-verifiable packs.
Negative tests
The implementation must refuse what the specification forbids: sealing incomplete chains, skipping the human gate, silently mutating sealed objects.
Adversarial tests
The implementation is attacked along the threat model: tampering, substitution, resealing abuse, cross-tenant attribution. Detection is mandatory, not best-effort.
One reference implementation currently exists: PREVORN, operated in production on Swiss infrastructure. Independent implementations are the explicit goal of publishing this framework, and the conformance plan is written so that we do not get to grade our own homework forever.
What PoG is today, and what it has not earned yet.
This table is the honest state of the framework. It will be updated as gates are passed, and only then.
| Stage | Gate | Status |
|---|---|---|
| Technical credibility | Evidence sealing, verification flow, event-to-evidence chaining, tenant-safe attribution, first conformance tests | IN PLACE · reference implementation |
| Framework formalization | Specification, threat model and conformance plan drafted, versioned and published for review | IN PROGRESS · drafts v1.0 |
| External review | Substantive adversarial review by parties independent of Smart STB | OPEN · reviewers wanted |
| Independent implementation | At least one conformant implementation not written by the framework authors | NOT STARTED |
| Protocol-grade language | Stable spec, provable conformance, coherent multiple implementations, survived adversarial review | NOT EARNED · by design |
Yes, this domain is called pog-protocol.org. The name states the destination, not the current status. Until the last gate above is passed, PoG is a framework, and our own claims policy forbids us from calling it anything more.
Rules we hold ourselves to.
- AuthorshipInitiated and maintained by Smart STB Sàrl, 17 Rue Dancet, 1204 Geneva, Switzerland. Editor: the PREVORN engineering team. No consortium is implied or claimed.
- Claims disciplineExternal claims about PoG are governed by the published Claims Policy. Staged language: framework today, specification plus conformance next, protocol-grade only if earned.
- Review processAnyone may submit a review by email. Substantive issues are answered publicly in the changelog. Silence is not an option we allow ourselves on documented flaws.
- VersioningDocuments are versioned explicitly. Breaking changes to sealed-artifact semantics require a major version and a migration note.
- LicensingSpecification documents are published under a license permitting free reading, citation and implementation. The PREVORN product code is not part of the framework and remains proprietary.
- No pay-to-playConformance language may never be bought. If a certification program ever exists, its tests will be the published ones, executable by anyone.