Model-based systems engineering
Model-based systems engineering, from requirements to verified variants
Ask a systems engineer what MBSE bought their project and the answer is often a good model and a tool nobody outside the team opens. The model is one island. Simulation results, test evidence, supplier data and decisions live elsewhere, and the thread breaks at the tool boundary. Ref is built for the boundary.
The situation
Where MBSE stalls
CATIA Magic, Rhapsody, Enterprise Architect and Capella model systems well. The trouble starts where the model has to meet evidence. A requirement is verified by a test that ran on a rig on a Tuesday. The link between the SysML element and the measurement file is a spreadsheet, and the spreadsheet is maintained by hand until it is not. MBSE promised traceability and delivered a model that a person has to reconcile with reality.
In Ref
A platform beside the modelling tool
Ref is not a SysML editor. It is the place where the model, the artefacts derived from it and the evidence produced against it stay connected. Requirements come in and go out as ReqIF. Architecture is structure: components, ports, connectors, interfaces and buses in a graphical editor. Behaviour is run: a Python or Modelica model against a parameter set, with the result recorded next to the requirement it verifies. FMI and SSP cover co-simulation.
Model files from your existing tool are stored with their versions and with what depends on them. Nothing forces the team to leave the editor it knows.
The cycle
The V-cycle as a by-product
Requirements and architecture on the left leg, integration and verification on the right, and the links between them created by doing the work. A run that verifies a requirement is linked because it was started for that requirement. There is no separate documentation task. A variant is compared against the criteria you set, and the decision is recorded with the evidence that was on the table.
Limits
What is missing, dated
No state machines, no parametric diagrams, no sequence diagrams; where you need behaviour, you run it. No XMI. SysML v2 textual export is planned for this year and not shipped as of September 2026. Sign-off records on a frozen version are on the roadmap and not shipped. If your process needs approval routes that block until signed, we are not ready for you.
If full diagram coverage and a SysML v2 pilot implementation are the requirement, CATIA Magic or Eclipse SysON fit better. We would rather you know that on this page.
Small teams
Without a method office
Ref runs without a method office. Start with the artefacts you have: a requirements sheet, a model directory, a folder of test results. Work under ISO 26262 or IEC 61508 needs the evidence chain this page describes. Ref provides the chain. It does not provide tool qualification, and holds no certificate for it as of September 2026.
Questions
Asked before every pilot
Which MBSE tools work with Ref?
Any tool that exports ReqIF for requirements. Model files of any tool are stored with their versions; the editor stays yours.
Does Ref replace our requirements tool?
It can hold the requirements. If Polarion or DOORS stays the master, ReqIF round-trips in both directions.
SysML v2?
Textual export of the architecture model is planned for this year and not shipped as of September 2026. Import follows export.
Requirement quality checks against INCOSE rules?
On request through the agent. There is no standing rule set with a side-by-side rewrite as of September 2026.
Can an agent write the SysML model from the requirements?
It drafts structure and a person corrects it. Peking University’s SysMBench, 2025, tested 17 models on 151 scenarios; the best scored 62 percent on the authors’ metric, and the authors call that poor. Those were the models of mid-2025 and the next generations do better; a person’s correction stays cheaper than an error nobody caught.
Related
If your model and your test evidence live in different places, a pilot on one subsystem costs nothing. Tell us about your project.