Open draft standard · AEL v0.1
Agent Evidence Level
An open draft standard for grading AI-agent evidence by what an independent party can verify and what omission they can detect.
- Status
- Draft v0.1
- Scale
- AEL-0 to AEL-4
- Specification
- CC BY 4.0
- Code
- Apache 2.0
The scale
Each level closes another omission gap
- AEL-0Authentic and ordered
- AEL-1Gaps and truncation evident
- AEL-2Cross-domain omission evident
- AEL-3Externally anchored
- AEL-4Counterparty confirmed
Producer or operator run
A signed evaluation package can't support a public numeric grade.
Independent verifier run
A signed verification record binds the grade to the exact run and evidence.
Maintained by PipeLab. Governance isn't independent yet. PipeLab also builds Pipelock and can't award AEL grades.
01 · Evaluation and verification
Running the checker isn't the same as earning a public grade
A producer or operator can run the checker and publish a signed evaluation package. That package displays EVALUATED. It has no grade or verifier field.
A public grade needs a signed verification record from someone who is neither the producer nor the operator. The record ties the grade to the exact run, artifact bytes, checker, corpus, and replay inputs.
02 · The boundary
AEL grades the supplied evidence for one run
It doesn’t certify a product or prove that every event was recorded. Each level closes a specific omission path by adding another source of signed evidence. Organizational independence still has to be disclosed outside the cryptography.
03 · PipeLab's role
The draft is vendor-neutral. Its governance isn't independent yet.
PipeLab maintains the v0.1 draft and also builds Pipelock. PipeLab can’t award grades, Pipelock receives no special treatment, and product marks stay out of the normative specification.
04 · Read and run it