create_refmodel() no longer returns the created reference model (the prior return logic is now inside a triple-quoted string). This is a behavior-breaking change: callers will now receive None instead of the created model/id/dict. Make create_refmodel() return the result of save_refmodel(...) (and remove the triple-quoted block rather than leaving it as a stray string literal statement).

A few API/implementation issues in save_refmodel():\n- It uses mutable default arguments (tags={}, op_tags=[]), which can leak state between calls. Use None defaults and assign inside the function.\n- The first parameter is named ReferenceModel, which reads like the ORM class rather than an argument; it’s clearer as model_cls (or omit it and reference ReferenceModel directly if it’s always the same).\n- Consider decorating with @db_session (like create_refmodel) so save_refmodel() is safe to call directly outside an existing session.\n- orm_create(..., fmt=fmt) is potentially problematic if orm_create doesn’t accept fmt (the prior code handled formatting after orm_create). If orm_create doesn’t support it, remove fmt=fmt and keep formatting in save_refmodel().

This file is named like a unit test but behaves like a destructive integration script: it performs DB reads/writes, deletes existing models, and has no assertions. This will be flaky and potentially harmful in CI/dev environments.\n\nSuggested direction (pick one):\n- Convert it to a real automated test (e.g., pytest) with assertions and isolated DB/fixtures (and avoid deleting arbitrary existing models), or\n- Move it to a scripts/ or examples/ location and keep it out of the test discovery path.\n\nAlso, the call uses tag={} but save_refmodel() expects tags=... (per its signature).