Topics discussed:

• The call opened on table formats. Delta, Iceberg and Hudi are all a metadata layer wrapped around Parquet — Delta adds a transaction log, so you can update a table repeatedly and time-travel back to earlier versions, and it supports merge semantics. The catch is that the transactionality is per table; there are no multi-table transactions. The whole category sits in the gap between dumping files on storage and running a relational database.
• The tagging work landed a structure on the spec: reserved tags now align with profiles — shareable, tabular, experimental — so each tag links to the profile it refers to. The FHIRPath subset moved into the shareable profile, and the test report groups results under those headings, with the boundary functions and join under experimental as things intended for the required subset later but not part of the normative release.
• Asked why the FHIRPath subset isn't stated in the ViewDefinition part of the spec, the answer was that it mirrors FHIR's own model. Base Observation is generic and permissive; to actually use it you profile it down to a sensible set of codes and values, which keeps the base stable and lets you tailor per use case. Likewise here: any FHIRPath is fine in a ViewDefinition, but the subset is what makes a view portable to another implementation, so it belongs in the profile that says so.
• A pull request stripping the metadata fields the profile copies from ancestor resources was closed rather than merged. The duplication exists because ViewDefinition is a logical model, not a real FHIR resource; if it is ever promoted, those fields should arrive by inheritance instead. Also noted: the functional model page is orphaned, with nothing in the spec linking to it.
• Around ten ViewDefinition examples for common resources still need converting to FHIR Shorthand. The upside is that the IG build validates them against the ViewDefinition logical model as part of the build, so mistakes like a string where a code belongs get caught.