I used AI tools to help draft and edit this post. I checked the facts and the wording, and the original lives at geekindata.com . The diagrams are on that page.

Canvas as surface, Dataverse as engine

Key takeaways

Prefer focused screens and a navigation model over one mega-screen that owns the whole process.

Design for delegation first; treat blue underlines as architecture debt, not editor noise.

Stop N+1 gallery lookups; prefetch with shaped Dataverse views and argue from Monitor traces.

Put eligibility, aggregation, and consistency in Dataverse; keep Power Fx for interaction.

Budget concurrent opens (many users at the same moment), not only cold start on one laptop.

Keep identity, Dataverse privileges, and ALM packaging in the same conversation.

Canvas is wonderful until one screen becomes a distributed system: nested galleries, OnStart fan-out, and business rules living in Power Fx because it was faster.

Microsoft's own performance reference architectures push a clear direction.…