AB-410 deep review — September 28, 2026¶
The AB-410 guide was read in full and mapped to 48 detailed objectives across seven groups. It now has six worked examples, ten labs, 48 answered checks and two focused Power Platform blog exercises. Eighteen offline assertions passed. SAP prompt-knowledge support remains an unresolved documentation gap.
This is same-context AI research and repair; independent human review is pending. No tenant, app, Power Fx runtime, Dataverse write, flow, approval, model, agent, deployment, billing or paid lesson was executed. The offline models do not prove concurrency safety, query delegation, service timing, model quality or authorization.
Exam and learning baseline¶
The official study guide retains its May 15, 2026 page date without a separate skills-effective date or announced revision. All 48 detailed objectives match the prior snapshot after whitespace normalization. The old snapshot abridged the audience section; the current accepted snapshot restores the complete source text and canonical heading and wrapping format. This is not evidence of a newly announced exam revision.
Original objective/status bytes are archived under
data/objective-snapshots/ab-410-2026-09-01-official-objectives.txt and the corresponding
status JSON. Historical review/audit references locate those archived bytes;
their original dates, hashes, outcomes and substantive evidence remain unchanged.
The credential still lists an active English, 120-minute exam and no Practice Assessment. The course lists three days and thirteen languages. Its four learning paths expose 17 modules: 3 + 3 + 7 + 4. Their earlier 15h26 total is historical; current fetched outlines do not expose those runtimes. The guide's 40–70-hour build-and-study budget is editorial.
The indexed public Udemy page shows Phillip Burton, a September update, 38 sections, 213 lectures and 21h09 total. Its description still says approximately 7h30 of AB-410 and promises removal of old PL-200 content from September 1. That inconsistency prevents treating the total as verified exam-only instruction. Direct automated access was blocked; the indexed public outline was available. Paid lessons and removal were unverified. Whizlabs and the partner/video pages returned content shells. Bounded exact-exam searches of Pluralsight, O'Reilly and MeasureUp found no offering; this is not proof of absence from the entire market.
Objective coverage¶
Every bullet has a hashed mapping in the operation record.
| Group | Count | Teaching and evidence |
|---|---|---|
| AI-enabled solution design | 5 | Requirements, built-ins, extensions, environments and ALM |
| Dataverse data models | 10 | Tables, columns, relationships, prompts, summaries, forms/views and access |
| Model-driven apps | 7 | Composition, generated pages, access, charts and dashboards |
| Canvas apps | 8 | Data, usability, automation, reuse, state, errors, monitoring and agents |
| Cloud flows | 6 | Triggers, connectors, approvals, actions, control and troubleshooting |
| Prompts and models | 8 | Contracts, apps/flows, knowledge, settings, inputs and model use |
| Business/process logic | 4 | Rules, process flows, derived columns and surface selection |
Material corrections and additions¶
- Freshness: add prompt-column state interpretation and distinguish stored output from a new successful generation. Explain summary scope, language, solution transport and Dynamics-specific table exceptions; correct the case scenario accordingly.
- Release and access: distinguish pipeline artifacts from data recovery, generated-page dependencies from editing history, authoring routes from regional limits, and bound record IDs from authorization. Test protected columns with nonadministrator personas and consider derived-output disclosure.
- App correctness: add incomplete-prefix examples, typed reuse and explicit error paths. A successful local query or type check does not establish complete data or valid business meaning; an error handler does not roll back prior effects.
- Automation: clarify update-column inclusion, action identities and operation keys. Separate interactive response deadlines from persisted business requests and long approvals.
- AI implementation: qualify older model references using current settings, distinguish per-call context from total interaction use, and explain saved JSON contracts, bounded knowledge and calibrated outcome measures.
- Licensing and business logic: preserve active-contract credit entitlements through their documented term, separate context-specific consumption, and add formula/rule/rollup timing and access boundaries.
Current implementation sources are linked beside the changed guide explanations. Long references were reviewed in the sections identified by individual source notes; retrieval was not treated as a complete audit of all linked procedures.
Unresolved evidence gap¶
The prompt-knowledge article names SAP in its connector instructions but omits it from the explicit support limitations. This review does not infer SAP support or exclusion from either sentence alone. The guide keeps that path unverified and schedules clarification. Other context, retrieval and relationship limits remain useful and documented.
Blog intake¶
Both accepted readings are bounded sections of long vendor roundups, not full article or linked-tool audits.
- Tiffany Treacy, September 17: the model app-builder, form-embedding and preview connector sections support an original 45–75-minute persona/artifact/binding review worksheet. No external coding tool, app creation or deployment was executed.
- Tiffany Treacy, May 14: the UDT section supports an original 30–60-minute type-versus-business-validation worksheet, checked against current App.Formulas guidance. No Power Fx runtime or linked code was executed.
These exercises supplement the blueprint; neither publication nor a GA headline establishes support for every associated preview feature or a particular tenant.
Verification and follow-up¶
Eighteen assertions cover prompt statuses, prefix recall, operation keys, response/approval deadlines, outcome denominators, review workload, stale totals and catalog arithmetic. They use synthetic values and do not simulate the actual Power Platform implementation. Labs remain learner exercises.
Three events recheck October 5 SAP support, October 12 feature/catalog boundaries
and November 1 credit transition. The shared learning catalog is synchronized
from the guide. Exact citation/fetch counts and executed repository/unit/site
checks are recorded in ADLC_Docs/operations/2026-09-28-ab-410-deep-review.json.
HTTP reachability, indexed visibility, full content access and semantic review
are recorded as distinct evidence.
All 48 guide citations are registered: 47 returned reachable HTTP responses, one direct commercial fetch was blocked, and none were classified missing/error. The Whizlabs, partner and video shells remain content-access limitations despite their successful HTTP responses.