Create a data standard to inform the network about your data
Before data can be discovered or used consistently across the Federated Learning Net network, it needs a shared description. That is the role of the data standard.
Why the data standard matters
The data standard is how your Client explains:
- which entities exist
- which fields are available
- how values should be interpreted
- which variables can be matched across sites
Without that layer, discovery becomes unreliable and federated analysis becomes difficult to reproduce.
What a good data standard should capture
At minimum, define:
- the main entities you want others to reason about
- the important variables for discovery and analysis
- expected types and units
- naming conventions that remain stable over time
Think of it as the contract between your local data reality and the shared network vocabulary.
Practical design advice
Start from real use cases
Do not model everything at once. Start from the questions the network should actually be able to answer.
Examples:
- Can others discover whether your site has a cohort with a certain diagnosis?
- Can a federated workflow expect the same lab values or outcome fields across participants?
Prefer stable semantics over local shortcuts
Local abbreviations may make sense internally, but they become confusing in a shared environment. Use names and definitions that remain understandable outside your institution.
Capture units and value expectations
Differences in units, coded values, or timestamp conventions are a common source of downstream errors. Write them down early.
Expect iteration
The first version of a standard is rarely the last one. Treat it as a maintained artifact rather than a one-time task.
After the standard is defined
Once the data standard is in place, the next steps are usually:
- configure connectors around that standard
- import representative datasets
- validate that discovery and mapping behave as expected
Continue with Add data to your Client.