Topics discussed:

• The connectathon surfaced a use case for FHIR servers using ViewDefinitions to denormalise the views applications need, since dashboards and reads are hard to build out of FHIR search. It wants nearly unlimited FHIRPath, including resolve(). The group's answer was to keep the subset as the required base and put resolve() and the rest into an optional advanced mode layered on top — an argument nobody expects to be over.
• A ViewDefinition runner is already in the IG publisher's code base. It is used for analytics over implementation guides themselves — counting code systems and codes — and on the International Patient Summary to compare profiles from different countries. Published IGs now carry a SQLite database of their content that anyone can query with a SQLite client.
• Making ViewDefinition an R6 resource was floated and pushed back on for timing reasons: R6 is far off and undated, and skipping straight to it would skip the stage that actually helps, where an incremental status gets the spec in front of more people and more use cases.
• The ballot-for-comment route — socialising the spec to people outside this call without committing to formal standards development — drew a pushback from FMG, who see too much going through that mechanism and expect this work to become an official standard regardless. The advice was accepted; the goal of getting in front of a wider audience did not change.
• MIMIC-IV on FHIR is the test vehicle: roughly 500 million resources of semi-public de-identified real inpatient data, rather than Synthea again. Pushing the full set through exposed validation problems and a transformation process that does not scale, so it is being rebuilt with validation in the loop. The functional model document is also close to done — implementers say it is the lens that makes the spec click, since you can read the four functions and port an implementation into your own language.