Skip to content

DP-600 deep review — September 27, 2026

The complete guide was read and 41 published October objectives in seven groups were mapped. July 21 remains the accepted baseline. The live page contains the October 19 capitalization change; historical snapshots and follow-up are preserved.

Learning improvements

Four worked examples teach additive OneLake grants, duplicate-dimension join totals, Direct Lake framing and weighted percentage totals. The guide now clearly separates OneLake's no-fallback behavior from SQL-based Direct Lake, and explains:

  • Supported engine/identity paths and sensitive model metadata.
  • Unenforced Warehouse keys and transformation quality checks.
  • Eventhouse batching/backfill/retention versus Import-model Delta exports.
  • Active relationship requirements for RLS and incremental-refresh deletion limits.
  • September result-set cache evidence and cloud/gateway ADBC differences.

There are ten labs and 44 answered checks. Six synthetic arithmetic/set assertions passed locally. No Fabric tenant, Power BI engine, SQL/KQL engine or driver lab was executed. Three ADBC research checkpoints were added for the planned October 2026, early-Q1 2027 and spring-2027 milestones; their review dates are not claimed vendor cutover dates.

Blog decisions

The guide adds two qualified readings:

  • DataZoe's April 21, 2025 Deep dive into Direct Lake on OneLake, in Microsoft's Power BI Updates Blog, for live model/report editing and architecture. Substantial available public text was reviewed via web search; direct automated retrieval of the migrated page was blocked. Initial-preview and early-access restrictions are reconciled with current documentation.
  • Chris Webb's April 5, 2026 Role-playing dimensions revisited, for design and version-specific investigation. The public article/discussion includes an April 9 author acknowledgment of a web-editor bug. It is not presented as a verified deployment recipe.

Neither article's screenshots nor tenant/model procedures were independently validated. Attribution, estimated reading time and original exercises appear in the guide.

Evidence and limits

ADLC_Docs/operations/2026-09-27-dp-600-deep-review.json records prior catalog data, guide hashes, objective hashes/positions and mappings, source observations, blog decisions, findings and validation. The objective count describes the live October version, not a newly accepted July snapshot.

Two paid books and the migrated Microsoft blog blocked direct retrieval. Public Learn path/module pages did not independently reverify the earlier 23-hour-20-minute duration total. No assessment, paid lesson or tenant workflow was completed. Preview, engine, identity, rollout and workaround limits remain explicit. Review and repair used the same AI context; independent and human review remain pending.

Publication uses unit-test, repository, strict site-build, generated-site and diff checks. The Microsoft review tracker keeps the remaining queue visible.

PL-300 follow-up

The subsequent PL-300 review clarified section 3 and answer 29: stored refresh-time columns describe the ordinary Import case. Other storage modes and user-context columns can evaluate at query time. The current calculated-column documentation was associated with DP-600, and the evidence preserves the guide hashes before and after this bounded correction. No engine execution was added; the publication gates cover the correction with the PL-300 change.