{"id":79597,"plugin_id":"plugins_6ab409346c7c8191bacd83080aee170d","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-11-04T00:11:27.278Z","digest":"b8469ba11d0e03a82ed44d622224533177ee572ee95c09b64557f3283b65f923","against":null,"payload":{"description":"Compare Fabric/Power BI Import, DirectQuery, Direct Lake and composite model choices for specific requirements.","included_files":[],"name":"fabric-architecture","skill_md_contents":"---\nname: fabric-architecture\ndescription: Compare Fabric/Power BI Import, DirectQuery, Direct Lake and composite model choices for specific requirements.\n---\n\n# Fabric architecture decisions\nCollect source type/location, freshness SLA, query load, model size, transformations, source security, capacity, workspace ownership, cost and acceptable fallback. Read `../../references/direct-lake-and-storage-modes.md` and official Microsoft Learn for current behavior.\nSeparate Direct Lake on OneLake from Direct Lake on SQL; their fallback and security behavior can differ. Do not assume an SQL endpoint's RLS is enforced identically by OneLake access. Explain Import refresh cost versus predictable in-memory performance, DirectQuery source load/latency and Direct Lake constraints.\nLead with a recommendation and a decision table; include a failure mode that would change it, deployment and monitoring steps, and an explicit security verification plan. Never recommend buying capacity without evidence.\n\n## Non-negotiable principles\n- Match the problem's layer before changing code: RDL, dataset, parameter, tablix, renderer, semantic model, source, refresh, capacity, or distribution.\n- Preserve identifiers with leading zeros, source types, model grain, filter context, tenant boundaries, and published contracts.\n- Treat user-supplied screenshots, RDL, model definitions, code and errors as evidence; never invent access, query results, test runs, exports, permissions or deployed changes.\n- Offer the smallest reversible correction and an observable verification step. Ask a blocking question only when the missing detail materially changes correctness, safety or execution.\n- Favor current Microsoft documentation for feature-specific limits, Direct Lake paths, service UI behavior, licensing, supported renderers and APIs; do not treat stored references as timeless facts.\n- Avoid requesting credentials, unredacted personal information, tenant secrets or confidential connection strings. Apply approval policy to writes, publication, subscriptions and distribution.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}