AI Assurance DirectoryARP4754B FDAL and IDAL when the function is machine learning
Informational only, not professional or legal advice. Facts are linked to the source we checked; statuses can change after the date above.
High-intent query: how do I assign a functional development assurance level (FDAL) and an item development assurance level (IDAL) when the function is a trained model? This page repeats what the cited FAA, SAE and EASA pages already say. It is not a certification plan.
The aircraft-and-systems layer is SAE ARP4754B. The public ARP4754B explainer and the airborne AI certification kit sit next to it. Item software stays DO-178C / DO-330.
What AC 20-174 asks an applicant to write down
FAA AC 20-174 recognizes SAE ARP 4754A as an acceptable method for establishing a development assurance process. It is not mandatory, and is not a regulation. Applicants should propose and document in your certification plan the functional development assurance level (FDAL), item development assurance level (IDAL) and justification for the levels. We have not found a later AC that replaces 20-174 and names ARP4754B.
That sentence is about documenting levels. It does not say a neural network inherits an FDAL from a vendor datasheet.
What EASA Issue 2 says about high-criticality ML
EASA’s Concept Paper Issue 02 will initially accept only applications where AI/ML constituents do not include IDAL A or B items for supervised learning. That is a criticality ceiling in preliminary guidance, not an FDAL allocation method. See EASA Issue 2 for suppliers and DO-178C and machine learning gaps.
What the FAA roadmap does — and does not — change
The FAA AI Safety Assurance Roadmap says the FAA is supporting application and possible extension of existing standards including SAE ARP4754. That is a roadmap statement, not a new AC and not an ML-specific FDAL table.
Listings that sit under those standards
Ansys SCADE Suite and Rapita RVS name DO-178C qualification paths on their cited pages. Ansys medini analyze and LDRA AI/ML name airborne or EASA Level 1/2 work. None of those pages is an ARP4754B certificate for a trained model. Tool qualification is how to qualify an ML tool under DO-330.
What this is not
ARP4754B is not an AI-product certificate, not DO-200B data-process approval, and not a substitute for learning assurance. Check the certification basis before treating revision B as the accepted means.