Certification Process

A review flow that ends in public proof.

The process turns an application into a decision, then turns the decision into a registry record and certificate page.

7
review steps
5
status states
Evidence
review base
Registry
public output
Expiry
renewal rule

Workflow

The review path should feel procedural.

AIAA needs a process page that reads like infrastructure, not marketing copy.

01

Application Intake

Collect owner identity, Agent name, category, country, product URL, and requested level.

Step
02

Identity Screening

Check owner consistency, public footprint, contact path, and registry eligibility.

Step
03

AI Assistance Declaration

Require disclosure of AI tools used, what AI helped with, what the candidate personally verified, and confirmation that the candidate can explain the submission.

Step
04

Evidence Review

Inspect logs, screenshots, workflow proof, documentation, repository or deployment context, and whether the evidence is reproducible.

Step
05

Capability Assessment

Review engineering judgment, debugging ability, reliability, task claims, scope limits, and evaluator notes.

Step
06

Decision

Issue approval, request information, reject, mark watchlist, or route to higher review.

Step
07

Publication

Create certificate page, registry row, status field, issue date, expiry date, and review summary.

Step

AI Assisted Policy

AIAA exams are AI assisted.

AIAA does not ban AI support. AIAA requires disclosure, verifiable evidence, and the ability to explain the final work in the candidate's own words.

Policy AreaExpectationFail Condition
Allowed toolsCandidates may use ChatGPT, Claude, Codex, Cursor, GitHub Copilot, and official documentation.Undisclosed AI assistance or hidden dependence on generated output.
Certification standardAIAA certifies engineering judgment, system execution, evidence quality, debugging ability, reliability, and production readiness.Memorization claims without runnable or reviewable evidence.
DisclosureCandidates must declare what AI was used, what AI helped with, what they personally verified, and confirm they can explain the submission.Missing AI Assistance Declaration.
Review modelAIAA uses evidence based review rather than invasive monitoring or browser lockdown.Prompt only, UI only, demo only, or non reproducible submissions.

Reviewer Checklist

What reviewers inspect.

The checklist helps applicants understand what evidence to prepare before review starts.

Review AreaEvidenceDecision Impact
Owner IdentityCompany records, owner contact, product URL, public profile.Controls whether a registry record is created.
AI Assistance DisclosureAI tools used, what AI helped with, what the candidate personally verified, and explanation confirmation.Controls whether the evidence package is reviewable.
Agent CapabilityTask examples, workflow logs, tool calls, supported use cases, debugging records, and reproducible output.Controls claimed scope and level eligibility.
Safety BoundaryPermission limits, fallback behavior, human override, prohibited use notes, and reproducibility constraints.Controls review conditions and warnings.
Operational ContextDeployment notes, monitoring, incident handling, organization proof.Controls Level 4 and Level 5 readiness.
Public PresentationWebsite claims, docs, labels, user facing promises.Controls registry wording and certificate text.

Decision Map

Every decision needs a next action.

No applicant should reach a dead end. Each state should route to review, update, renewal, appeal, or public record.

Pending
01
Verified
02
Info Needed
03
Expired
04
Revoked
05
StateNext ActionPublic Output
PendingWait for reviewer or submit missing details.Pending registry row when enabled.
VerifiedPublish certificate and registry record.Public record and certificate URL.
Information NeededApplicant uploads missing proof.Private review state.
ExpiredApplicant requests renewal.Expired status remains visible.
RevokedApplicant may appeal or submit remediation.Revoked status remains visible.

Next Action

Prepare an Agent review file.

Start with identity, ownership, product URL, capability claims, and evidence links.