Microsoft refresh group 3 of 5 — October 28, 2026¶
This group refreshes ten guides and maps 658 current detailed objectives. 18 guides in two groups remain in the frozen 48-guide refresh after this group. Two subset exam researchers read five complete guides each; technical and learning-resource researchers record their selected body-reading scopes. The coordinator owns shared edits.
The operations ledger records exact current guide/objective/teaching-span hashes, opened URLs, role findings, applied repairs and preserved history. This is bounded AI-assisted content review. Human review, a full independent rubric-2 audit and live product execution remain separate.
| Guide | Current objectives | Source gate | Links |
|---|---|---|---|
| AB-620 | 44 | Passed; human review pending | 59/61 reachable; 2 access-blocked; 0 missing/error |
| AB-731 | 39 | Passed; human review pending | 34/36 reachable; 2 access-blocked; 0 missing/error |
| AI-300 | 58 | Passed; human review pending | 59/61 reachable; 2 access-blocked; 0 missing/error |
| AI-901 | 29 | Passed; human review pending | 39/42 reachable; 3 access-blocked; 0 missing/error |
| AZ-104 | 82 | Passed; human review pending | 64/67 reachable; 3 access-blocked; 0 missing/error |
| AZ-140 | 78 | Passed; human review pending | 58/61 reachable; 3 access-blocked; 0 missing/error |
| AZ-305 | 49 | Passed; human review pending | 51/54 reachable; 3 access-blocked; 0 missing/error |
| AZ-400 | 86 | Passed; human review pending | 82/83 reachable; 1 access-blocked; 0 missing/error |
| AZ-700 | 136 | Passed; human review pending | 75/77 reachable; 2 access-blocked; 0 missing/error |
| AZ-900 | 57 | Passed; human review pending | 69/71 reachable; 2 access-blocked; 0 missing/error |
Objectives, teaching and history¶
All ten detailed objective snapshots remain byte-identical to the accepted baselines. AB-620 and AI-300 have no published skills-effective date: their study-page Last Updated metadata now says October 16 and October 15 respectively. Those two status metadata updates do not establish new exam-effective revisions. No objective archive or historical review-path migration is needed. Earlier source/deep/audit review values remain unchanged.
The initial complete semantic matrices record 130 partial named workflows: 38 across subset A and 92 across subset B. Forty-two original worked-example packets plus a same-team Site Recovery configuration follow-up address those gaps. Final semantic maps distinguish actual teaching from a generic heading or a valid hash. Eleven local Python AST/YAML syntax checks were performed by the exam researcher; they do not execute an SDK, cloud pipeline or service.
AB-620¶
Update undated page metadata, live course and sign-in Practice Assessment observations. Separate classic lessons from the harness-aware design module. Add connector knowledge/tool, connected-agent and pipeline-evaluation gate workflows with explicit identity, publication and failure boundaries.
AB-731¶
Add concrete Basic/Premium, per-app and mobile capability comparisons and a pilot-investment worksheet. Editing a spreadsheet, chatting over a file and broad organizational grounding have separate entitlement/data boundaries. Move readiness before the final Places to learn section.
AI-300¶
Update undated page metadata without inventing an effective date. Add infrastructure deployment, versioned command-job, model-card serving and embedding-adaptation workflows. Label upstream training as Corroborated — verify current and retain the managed-endpoint support gap. Clarify preview managed-network CLI versus REST/IaC account provisioning.
AI-901¶
Add voice portal, Speech SDK, vision-input, image-generation and asynchronous multimodal extraction contracts. Keep API surfaces and optional previews separate. Use supported minimum SDK requirements and record resolved versions. Add the 2,000-page asynchronous versus five-page synchronous completeness check and current Foundry IQ learning route.
AZ-104¶
Add hierarchy moves, file/share recovery, ARM/Bicep modifications, VMSS creation, DNS tests, two vault families, backup reports/alerts and full Site Recovery configuration/failover/commit/reprotect. Distinguish the same-region Azure Backup vault example from the outside-source-region Site Recovery vault. Proposed workflows are unexecuted.
AZ-140¶
Add profile storage/identity, Windows licensing, endpoint onboarding, ODFC/masking, redirection/printing, personal assignment/power-on and session-maintenance workflows. Corroborate the ODFC policy registry base with current documentation and official FSLogix 26.08 ADMX. Scope Government Shortpath endpoint preview separately from the broader Multipath GA matrix; elapsed AVD classic retirement does not retire AZ-140.
AZ-305¶
Add on-premises authorization and analytical-serving design decisions. Separate Front Door classic service retirement in March 2027 from migrated classic endpoint cutover in March 2028. A successful profile migration does not finish DNS/direct-hostname cleanup.
AZ-400¶
Add traceability, test/coverage YAML, templates, flags, guest configuration, self-service, security connectors and telemetry workflows. ADE remains exam scope despite announced February 22, 2027 retirement. Lockfiles cover reusable-workflow references; their internal dependencies use the called repository lock. The dated GA announcement and lagging preview label are both visible. Explicit missing-file/publication failure controls enforce required test evidence.
AZ-700¶
Add public-prefix/BYOIP, Route Server/Network Manager, posture/query, P2S, ExpressRoute, Virtual WAN and application-delivery workflows. Current non-AZ VPN gateway guidance distinguishes ongoing traffic/support from post-rollout management lockout. Scope control-plane/routing/security and actual forwarding tests separately.
AZ-900¶
Retain describe-level objective coverage, add a cost-decision worksheet, clarify commitment/refund boundaries from current primary terms and avoid promising unverified free-sandbox entitlement. No pricing commitment or purchase is required for the paper exercise.
Evidence and validation¶
The initial health pass checked 506 sources: 483 reachable, 23 automation-blocked and zero broken. All 86 additional associations and two targeted follow-up observations returned HTTP 200. Eighty-three new evaluated sources were registered. Thirty learning/exam catalog entries and selected technical notes were updated with their precise reading scopes. Public outlines, body reading, syntax parsing and live execution are different evidence; paid interiors, assessments and videos were not consumed.
Two fresh-context read-only reviewers read all ten complete final guides and diffs, checked all 658 detailed objectives and independently opened 43 selected primary/upstream URLs. Both bounded reviews passed after two medium AZ-400 repairs and one low AI-901 SDK-version wording correction. Reusable-workflow references and internal dependencies now have the correct lockfile boundaries; the YAML explicitly fails missing/unpublished test results. The earlier semantic matrices and reviewer findings remain preserved alongside the repair/recheck evidence.
All 199 tests and repository validation passed. Preservation checks retain all 309 earlier source-review objects, 186 deep-review receipts and 42 AI-audit batches unchanged, including a byte-identical AI-audit file and ten unchanged objective files. The operations ledger records website build/link validation after it runs; stage/commit/push requires those checks to pass. No guide or Markdown changes follow that website gate without another appropriate check. Bounded source passes do not complete human review, a full independent semantic audit or live labs; a maintenance-PR Pages build is not production deployment.