Confidence Levels
AGF labels every substantive claim with one of three confidence levels so readers can audit where AGF stands on evidence.
AGF is a synthesis framework. Some of what it claims is settled — multiple authoritative sources agree. Some is our own synthesis — plausible, single-source, not yet independently validated at scale. Some is flagged as open for investigation. Readers deserve to know which is which.
Every substantive claim across AGF content is labeled with one of three confidence levels. You'll see them inline next to claims and as block markers on sections where the whole topic is speculative.
The three levels
Established. Clear evidence; multiple sources agree. The claim is load-bearing and defensible against scrutiny. Most synthesis of NIST, OWASP, CSA, ISO, and EU AI Act material lands here because it represents the consensus of those communities.
Informed. Our synthesis; single-source; plausible but unverified at scale. This is where AGF does its composition work — connecting patterns across frameworks, naming relationships that haven't been formally named elsewhere. Credible but not yet validated by independent review or production deployment.
Open. Flagged but speculative; needs investigation. Claims marked Open are deliberately visible so readers, contributors, and reviewers can help close or refute them. Architectural commitments that don't yet have implementation artifacts also land here.
How they appear
Inline — a small badge next to the specific claim it labels:
Every primitive is derived from prior art.
Block — when an entire section, note, or artifact is at a single confidence level:
Why we do this
Promising rigor without showing it is an easy trap for reference frameworks. Naming every claim's confidence level does three things:
- Forces honesty at author time. You can't hide a stretch claim behind passive voice when the prose requires a label.
- Invites contribution. An Open claim is an explicit ask for evidence, critique, or a counter-example.
- Makes the framework auditable. Any reader, reviewer, or regulator can ask: "Which claims are Established? Which are Informed? Which are Open?" — and get a direct answer from the page itself.
Related
The same gradient applies to internal review findings. See docs/reviews/README.md in the source repository for how the gradient is used to label findings in AGF's own governance ledger.
The 19 Primitives
Named patterns for governed agentic systems — runtime, lifecycle, and substrate primitives organized into the Rings Model.
Governance Decision Record (GDR)
AGF's canonical audit artifact at gate boundaries — a single record format that serializes any Gate Resolution or Domain Outcome into an auditable artifact.