Official-source freshness scans

Source freshness is a discovery gate for official material that is not already represented in the library. It complements two narrower deterministic checks:

• the objective monitor detects changes on configured blueprint pages; and
• the source-health monitor checks reachability and public metadata for registered URLs.

Neither existing monitor can discover a newly published documentation page, release note, retirement notice, renamed feature, or replacement source. A freshness scan searches for those external deltas and records review work without silently changing a guide.

Operating boundary

• Scan public, first-party vendor sources only. Do not use recalled questions, dumps, confidential training, customer information, or authenticated material. The preparer enforces public access, authority classes 1–3, and an official source type; paid, partner-restricted, third-party, and expert resources are excluded even if their existing catalog rank is unusually high.
• Run read-only discovery before repair. The scanning agent reports evidence; a separate review decides whether to queue, apply, dismiss, or block a finding.
• Complete that review in the same work wave when the evidence supports a bounded guide note with reasonable confidence. A queue is a short decision boundary, not a parking place for verified concerns that can already help a learner.
• Prefer vendor-sized batches of 10 guides and never exceed 12.
• Treat search snippets as discovery hints, never as evidence. Open the canonical official page before recording a finding.
• Do not infer product behavior from an exam blueprint or course landing page.
• Do not mark a guide community reviewed. An AI freshness scan is not human review.

Required search channels

For every selected guide, inspect all applicable channels:

1. The official credential page and current exam blueprint, including effective dates, beta status, replacements, and retirement announcements.
2. Official product documentation for each volatile or weakly supported objective.
3. Official release notes, roadmaps, changelogs, retirement notices, and product announcements.
4. The current data/sources.json and data/source-candidates.json entries for the exam so existing evidence is not rediscovered as new.
5. The guide's VERIFY CURRENT boundaries and unresolved audit findings, looking specifically for newer first-party evidence or contradictions.

Record the exact canonical URL, visible title, dated change when the vendor publishes one, affected exams, catalog status, evidence, confidence, and a bounded recommended action. If the official source does not publish a date, say so rather than guessing.

Dynamics 365 and Power Platform release discovery

From September 2026, use the AI at Work roadmap for new Dynamics 365, Power Platform and Dataverse disclosures. The older Learn release plans remain useful historical evidence. Microsoft's transition announcement announces November 15, 2026 retirement of Release Planner; verify that event after its date rather than describing it as already completed.

For an applicable feature, record its ID, product, status, estimated date and source observation date. Confirm behavior and support boundaries against current product documentation and, where required, the tenant's release state. Roadmap or blog wording alone does not establish GA, licensing, regional availability or exam scope. Carry material changes into a dated maintenance event and the certification's next review. The scheduled repository workflow detects and queues review work; it does not ingest the roadmap's RSS/CSV or perform this semantic assessment automatically.

Five required checks

Each result records passed, finding, blocked, or not-applicable, plus evidence notes for:

1. Official blueprint and lifecycle: scope, effective dates, beta/GA state, replacement, and retirement signals.
2. Official product documentation: newly published or materially revised implementation evidence for tested capabilities.
3. Official release channels: relevant release notes, roadmaps, changelogs, retirement notices, and announcements.
4. Catalog comparison: every reported URL is compared with approved sources, healthy redirect/canonical aliases, and queued candidates.
5. Contradiction and gap review: newer evidence is checked against guide claims, VERIFY CURRENT statements, and unresolved audit findings.

Outcomes and dispositions

• current: no review-gated or blocked findings remain. Historical findings may be retained as applied or no-action with a resolution.
• review-required: one or more findings are queued for source or guide review.
• blocked: an applicable official channel could not be checked reliably.

A queued finding must point to a candidate in data/source-candidates.json. An applied finding must point to an approved source in data/sources.json and state what was changed. no-action preserves useful evidence that did not warrant a catalog or guide change. blocked records the exact access or evidence limitation.

A blocked official channel does not end the research pass. Check vendor-owned code, examples, training, and announcements, then maintained upstream material, named experts, and independent implementation reports under the source-quality hierarchy. Add the strongest useful provisional model to the guide with a dated Corroborated — verify current or Unconfirmed — validation needed label, exact citations, the unresolved official-evidence gap, and a safe recheck path. The official freshness result remains blocked until that channel is auditable, even though the guide has been improved.

Preparing a scan

For research-heavy refreshes, use the subagent refresh workflow. Research findings remain read-only until the coordinator evaluates and applies them; delegation does not change this rubric or establish independent human review.

The preparer packages current exam metadata, the guide-bound source baseline, official catalog entry points, registered first-party sources, health states, and queued candidates. Explicit exam codes are useful for a curated wave:

python scripts/prepare_source_freshness_scan.py \
  --batch-id freshness-2026-09-05-github-microsoft-azure \
  --exam-code GH-100 --exam-code GH-200 --exam-code GH-500 --exam-code GH-900 \
  --exam-code AB-100 --exam-code AB-650 --exam-code AB-900 --exam-code MD-102 \
  --exam-code AI-103 --exam-code AZ-700 --exam-code AZ-800 --exam-code AZ-802 \
  --output .site-build/freshness-2026-09-05.json

For recurring work, omit the exam codes. The default queue selects up to ten guides whose current baseline has never been scanned or was last scanned at least seven days ago, ordered by lifecycle and source-volatility risk. Use --min-age-days to change the cadence and --vendor-id to constrain a provider.

The baseline hash covers the current guide, exam lifecycle metadata, and registered authority-class 1–3 sources. A relevant guide or official-source catalog change makes the prior result ineligible as the current baseline. Time still matters: unchanged local inputs become due again because the external web can change. Completed batch hashes are immutable historical evidence: a later local change makes that baseline due again, but does not invalidate the earlier record. Blocked or future-dated results never suppress recurrence.

Review and completion

1. Run the deterministic objective monitor for the selected exam codes.
2. Give the generated manifest and this rubric to a fresh-context agent.
3. Verify every reported change against the opened first-party page.
4. Add genuinely undecided discoveries to the candidate inbox. Otherwise evaluate and promote useful sources, add the bounded concern or provisional explanation to the living guide, and record the applied source/content change during the same repair wave.
5. Record the completed results in data/source-freshness.json; never delete an earlier finding to make the summary pass.
6. Run repository tests, source checks appropriate to changed URLs, and generated site validation before merging.

Run volatile GitHub, Microsoft 365, Copilot, Power Platform, and Azure batches weekly. A monthly cadence is usually sufficient for slower, stable providers, with an immediate manual scan after an official exam or product announcement.

Initial GitHub and Microsoft/Azure scan

The September 5, 2026 baseline ran three independent, fresh-context agents across all five existing GitHub guides and the eight recently repaired Microsoft platform/Azure guides. A deterministic live check also confirmed that all 13 configured objective and status snapshots were unchanged.

Batch	Exams	Current	Review required	Queued guide/source impacts	Applied findings
GitHub	5	1	4	15	3
Microsoft platform	4	0	4	10	2
Azure	4	0	4	8	0
Total	13	1	12	33	5

The applied findings correct GH-300's retired github.com Spark experience, Copilot CLI/app content-exclusion support, and public-preview Copilot approvals, recognize a redirected Copilot Studio source's canonical harness overview, and synchronize AB-650's current Microsoft 365 and AI Services display name. The candidate inbox contained 23 unique first-party URLs for separate relevance and content review at scan time. That count was lower than 33 because a single source could affect multiple guides. All candidates were reviewed by September 7; the inbox is now empty.

The scan also found the official GH-600 study guide. On September 7, that expansion was approved and completed with a substantive beta guide, objective and status snapshots, source-validation evidence, official learning paths, labs, and explicit product-volatility warnings. No public Microsoft implementation documentation for SSH Direct was found, so the existing AZ-800 and AZ-802 evidence boundaries remain in place.

The two remaining GitHub candidates were also applied to GH-100: Enterprise Live Migrations is now bounded to its supported GHES-to-GHEC Data Residency path, and the GHES release ledger is the current support/compatibility check. Cisco's CCNA credential page still conflicts with its dedicated exam page on price and language; both are cited, the narrower exam page is used for planning, and checkout verification remains required.

A separate September 6 best-effort follow-up expanded beyond the official-source-only freshness boundary and found Microsoft-owned code, Microsoft Press, upstream packaging, and expert evidence for a provisional Hyper-V/VSOCK/OpenSSH model. Those exact URLs are queued as candidates and summarized in SSH Direct evidence and validation boundary. This does not rewrite the September 5 scan result or unblock source validation. The same follow-up now includes separately bounded Debian, Ubuntu, RHEL, CentOS Stream, Fedora, SLES/openSUSE, and Oracle Linux observations. Cross-family evidence is useful for forming and testing the technical model, but it is not a substitute for an exact host/guest/version result or a vendor support statement.

Repository-wide freshness expansion

On September 7, 2026, the risk-ordered queue contained 192 current guide baselines without an eligible rubric result. Batch 04 reviewed the first 12 guides against live first-party blueprint, product-documentation, release, and catalog evidence.

Batch	Exams	Current	Review required	Blocked	Queued findings	Applied findings revalidated	No action
04	12	5	5	2	17	26	1
05	12	9	3	0	13	1	0
06	12	10	2	0	6	0	3
07	12	12	0	0	0	0	5
08	3	3	0	0	0	0	3
09	4	4	0	0	0	0	0
10	5	5	0	0	0	0	0
11	5	5	0	0	0	0	0
12	11	11	0	0	0	0	10
13	6	6	0	0	0	0	6
14	12	12	0	0	0	0	12
15	5	5	0	0	0	0	5
16	12	12	0	0	0	0	1
17	10	10	0	0	0	0	0
18	3	3	0	0	0	0	3
19	6	6	0	0	0	0	5
20	4	4	0	0	0	0	3
21	8	8	0	0	0	0	8
22	7	7	0	0	0	0	5
23	7	7	0	0	0	0	4
24	6	6	0	0	0	0	1
25	11	11	0	0	0	0	21
26	9	8	1	0	1	0	0
27	7	6	0	1	0	1	0
28	6	1	0	5	0	5	0
29	10	7	0	3	0	3	3

The two blocked results are CISM's announced November 2026 outline, which is not yet publicly available in sufficient detail, and Fortinet's MSSP credential, whose official page still says only Coming soon!. Blocked results intentionally remain due. The other ten results reduce the current-baseline queue to 182 guides. The no-action finding closes the freshness discrepancy for the CCNA credential page: its formerly conflicting price and language claims are no longer visible, so the current exam page remains authoritative.

Batch 05 covers 12 Microsoft guides. All live objective and status snapshots were unchanged. AB-100, AB-620, and PL-900 retain review-required outcomes for 13 exact first-party candidates covering the AI at Work roadmap transition, current Copilot Studio harness documentation, Agent 365 lifecycle evidence, Power Platform deprecations, and Microsoft 365 Copilot release notes. The other nine guides are current. This reduces the current-baseline queue to 170 guides.

Batch 06 covers 12 more Microsoft guides. AB-410 and PL-300 retain six shared roadmap, harness, and deprecation candidates; the other ten results are current. Three deterministic monitor alerts were manually resolved as no-action extraction drift: PL-300 and AB-730 changed wrapping or audience prose without changing domains, weights, or tasks, while AB-410's extractor omitted its audience preamble but retained the same objective groups and bullets. The current-baseline queue is now 158 guides.

Batch 07 covers 12 more Microsoft guides and found no actionable freshness gap. Five raw monitor alerts were resolved as no-action representation drift after manual comparison. AB-731 and AB-250 gained expanded audience or skills-at-a-glance rendering; MB-230, MB-330, and MB-500 expanded previously condensed objective text. Their published skill dates, domains, weights, and assessed tasks remain substantively unchanged. The current-baseline queue is now 146 guides.

Batch 08 closes the remaining Microsoft provider queue with MB-310, MB-800, and MB-820. Each raw monitor alert is a condensed-snapshot versus expanded-live-text difference; the published skill date, domain weights, and assessed tasks are unchanged. All three results are current, reducing the current-baseline queue to 143 guides.

Batch 09 covers CKAD, CKS, LFCA, and LFCS. All live objective and status checks are unchanged. CKAD and CKS remain on Kubernetes v1.35, while the already registered v1.37 release watch and Ingress NGINX controller-retirement boundary remain correctly bounded. All four results are current, reducing the queue to 139 guides.

Batch 10 covers the five remaining CompTIA baselines: Cloud+, Network+, Security+, Linux+, and Tech+. Their live exam versions, weighted domains, delivery contracts, and published or estimated retirement watches are unchanged. All five results are current, reducing the queue to 134 guides.

Batch 11 covers all five Red Hat guides. Their live objectives and product-version contracts are unchanged: RHEL 10 for EX200, OpenShift 4.22 for EX280, OpenShift AI 3.3 with OpenShift 4.20 for EX267, Quarkus 3.8 for EX378, and the current purchasable version boundary for EX294. EX280 still requires checking the assigned LMS version. All five results are current, reducing the queue to 129 guides.

Batch 12 covers 11 Fortinet guides and excludes the same-day MSSP blocker. Ten monitor alerts were resolved as no-action composite-extraction differences: the entry guides combine credential and course evidence, solution tracks combine several exam pages, and the SASE page lists its retiring 7.6 block before its current version 26 block. Manual review also confirmed that the announced FortiMail WorkSpace, FortiVoice, FortiAnalyzer, FortiRecon, and FortiDeceptor pages still say Coming soon!. The guides already preserve those gaps. All 11 results are current, reducing the queue to 118 guides.

Batch 13 closes the six remaining publishable Fortinet tracks. Manual comparison confirmed their live versions, delivery contracts, domains, and tasks; the monitor hashes differ because its reduced extraction omits snapshot detail, and NSE 6 SASE uses a composite three-exam snapshot. All six alerts are recorded as no-action representation findings. The results are current, reducing the queue to 112 guides.

Batch 14 covers 12 Palo Alto Networks guides. The deterministic monitor cannot parse the official PDF datasheets, and the concise Apprentice HTML objective falls below its length threshold. Browser-visible first-party evidence still exposes the expected credential identities, roles, and weighted blueprints; current source-health records remain healthy and no lifecycle delta or contradiction was found. The 12 format-limit alerts are no-action findings. All results are current, reducing the queue to 100 guides.

Batch 15 closes the remaining five Palo Alto Networks guides. Their current first-party PDF datasheets remain browser-readable and aligned with the guide-bound roles and weighted blueprints, while the HTML-oriented monitor reports format errors. Those five errors are recorded as no-action format-limit findings. All results are current, reducing the queue to 95 guides.

Batch 16 covers all 12 Oracle guides. Eleven extracted objective and status pairs are unchanged. Java 1Z0-830 adds introductory, audience, and closing preparation prose while preserving its Java SE 21 objectives, exam code, and 120-minute contract; that alert is a no-action content-boundary finding. All results are current, reducing the queue to 83 guides.

Batch 17 covers the ten remaining AWS baselines. Every live AWS exam guide matches the committed credential identity, objective domains, weights, and status snapshot. No replacement, retirement, contradictory service boundary, or uncataloged first-party source was identified. All ten results are current. A complete baseline-hash recount after this batch found 85 guides still due, not 73: the earlier running subtraction did not account for 12 baselines invalidated by intervening guide or source-catalog changes.

Batch 18 covers Cisco 100-150, 100-160, and 200-901. The two CCST objective hashes are unchanged; their alerts only initialize previously absent delivery-status hashes. The CCNAAUTO landing-page extraction changed with Cisco's credential rename, but the guide already uses the detailed official v1.1 blueprint and records the landing page's v1.0 label as a first-party discrepancy. All three alerts are no-action findings and all results are current, reducing the recomputed queue to 82 guides.

Batch 19 covers all six C and C++ Institute guides. CPE-20-01 is unchanged. Five other live pages expose expanded instructional wording where the committed snapshots use concise paraphrases, while preserving the same active exam codes, July 2025 alignment dates, block weights, topic order, and numbered objectives. Manual objective-by-objective comparison found no scope or lifecycle delta, so those alerts are no-action representation findings. All six results are current, reducing the queue to 76 guides.

Batch 20 covers all four JS Institute guides. JSE-40-01 is unchanged. The other three live pages now label their detailed objective section Exam Syllabus Contents, while the HTML adapter still expects the former Exam Scope heading. Manual extraction confirmed the active codes, block weights, objective counts, and detailed topics match the normalized guide snapshots. The three errors are no-action adapter-limit findings; all four results are current, reducing the queue to 72 guides.

Batch 21 covers the eight remaining Python Institute guides. Six live pages expose full provider wording where committed snapshots use normalized paraphrases; their codes, status, update dates, block weights, item counts, and topics remain aligned. PCET-30-01 and PCAT-31-01 remain browser-readable, but their pages no longer expose the download marker used as the monitor's end boundary. Manual extraction found no content delta. The PCAP, PCPP1, and PCAT successor watches remain disclosed in their guides with no newer effective scope. All eight alerts are no-action findings and all results are current, reducing the queue to 64 guides.

Batch 22 covers the seven remaining Google Cloud guides. Cloud Digital Leader and Generative AI Leader are unchanged. The other five objective hashes are also unchanged; their alerts initialize previously absent delivery-status snapshots. Existing branding notices for Associate Cloud Engineer, Professional Cloud Architect, and Professional Machine Learning Engineer are already explicitly disclosed in the guides, including the Gemini Enterprise transition where applicable. The five alerts are no-action baseline findings. All seven results are current, reducing the queue to 57 guides.

Batch 23 covers all seven Databricks guides. Five objective/status pairs are unchanged. The Generative AI Engineer alert is only hyphen-to-en-dash typography plus status initialization, and the Apache Spark Associate alert only initializes status. A blocked historical Spark 3.0 community notice and a blocked supporting MLflow Optuna tutorial are recorded separately; accessible current blueprints and adjacent official product documentation still support both assessments. These four items are no-action findings. All seven results are current, reducing the queue to 50 guides.

Batch 24 covers all six IBM guides. Five objective/status pairs are unchanged. C1000-185 differs by one trailing space after its title; every status, delivery, weighted-section, and objective line remains identical. That alert is a no-action whitespace finding. All six results are current, reducing the queue to 44 guides.

Batch 25 covers all 11 Splunk guides. SPLK-5001's live HTML objective and status snapshots are unchanged. The other ten official blueprints are PDFs that the HTML monitor cannot parse; local extraction with the already-installed PDF reader confirmed every live weighted-domain sequence exactly matches its committed snapshot. Each guide also has one or more supporting Splunk Help or community URLs that block automated health access. The ten format limits and 11 grouped supporting-access limits are recorded as 21 no-action findings, while accessible blueprint and credential evidence supports every assessment. All 11 results are current, reducing the queue to 33 guides.

Batch 26 covers GH-100, three HashiCorp guides, and five Microsoft Office guides. Eight objective/status pairs are unchanged. The Vault Operations page is readable and its 130 objective lines are unchanged, but HashiCorp now labels the credential Vault Operations Advanced instead of Vault Operations Professional. That real title conflict is queued against the official certification catalog for separate metadata, guide, and adapter review; it was not silently repaired during discovery. Eight results are current and one is review-required, reducing the rubric queue to 24 guides.

Batch 27 covers four ISC2 and three ISACA guides. The four ISC2 outlines and the CISA and CRISC outlines are unchanged. CISM's current four-domain outline is also unchanged and still announces a November 3, 2026 update, but its official update URL continues to return only a JavaScript application shell without auditable future domains, weights, or objectives. That known limitation remains blocked rather than being guessed. Six results are current, reducing the queue to 18; CISM remains due.

Batch 28 covers four Salesforce and two ServiceNow guides. The Salesforce Platform Administrator Trailhead objective and maintenance snapshots are unchanged. The other three Salesforce Help pages return only Loading and a CSS error, while both ServiceNow pages return only a title plus Loading... to the deterministic and available browser checks. HTTP success alone is insufficient to claim their weighted blueprints are unchanged, so all five are explicitly blocked. One result is current, reducing the queue to 17; the five blocked results intentionally remain due.

Batch 29 covers four Snowflake, three MongoDB, and three NVIDIA guides. NVIDIA's three objective/status pairs are unchanged. Three active Snowflake objective hashes are unchanged and only initialize status; retired SOL-C01 remains correctly bound to its registered retirement and COF-C03 replacement evidence. The three MongoDB University study-guide and credential URLs resolve only to a generic client-rendered course shell, so current weighted scope cannot be re-audited and those results are blocked. Seven results are current, reducing the queue to 10; the three MongoDB results intentionally remain due.

The September 7 closeout recomputed every baseline hash rather than relying on the running subtraction. Before remediation, all 222 guides had a rubric result for their current local baseline: 188 current, 24 review-required, 10 blocked, and zero missing.

Batch 30 revisits the four AWS review-required guides under the living-guide rule. Nine official candidates were promoted and health checked. Ten per-guide findings were applied across AIB-C01, ANS-C01, MLA-C01, and MLA-C02: exact AI Business Strategist scope/exclusion appendices; Transit Gateway policy-based routing and its Reachability Analyzer boundary; the MLA-C01/C02 January 14, 2027 transition; and SageMaker monitoring availability changes. Each guide now separates confirmed facts, useful adjacent product context, and the concern that still needs rechecking. All four results are current, leaving 192 current, 20 review-required, 10 blocked, and zero missing on the post-batch baseline. The blocked results remain visible even when supplementary evidence improves their guides.

Batch 31 applies the same rule to GitHub. Seven official changelog sources were promoted and health checked, and 13 per-guide findings were applied across GH-100, GH-200, GH-300, and GH-500. The guides now cover the September runner and Node deadlines, expanded Actions retention, reusable-workflow identity and least-privilege permissions, Copilot policy/retention/billing and model changes, and CodeQL 2.26.4. Each note distinguishes durable governance or security lessons from volatile release details and from the unchanged exam blueprints. The three review-required guides are now current, leaving 195 current, 17 review-required, 10 blocked, and zero missing.

Batch 32 resolves HashiCorp's Vault Operations credential rename. The official catalog, exam-content page, learning path, and orientation now consistently say Vault Operations Advanced; a validated comparison found all 130 objective lines unchanged apart from the title. Display metadata and canonical URLs now use Advanced, while the stable internal VAULT-OPERATIONS-PROFESSIONAL identifier is retained and explicitly documented because HashiCorp has not published an unambiguous replacement exam code. The result is current, leaving 196 current, 16 review-required, 10 blocked, and zero missing.

Batch 33 reviews 12 Microsoft platform, Azure AI, endpoint, administration, and Power Platform guides under the living-guide rule. Seventeen queued first-party sources were promoted and health checked, including the AI at Work transition, Copilot and product release feeds, Agent 365 GA, Power Platform and Dynamics deprecations, Intune changes, Entra/Defender/Purview feeds, Content Understanding changes, and AI-500 lifecycle and migration evidence. Each guide now records the concern and distinguishes blueprint scope from roadmap estimates, preview/GA state, staged rollout, retired behavior, and independent learning aids. AB-620's obsolete two-experience table was replaced with the current three-harness model. All 12 results are current, leaving 208 current, four review-required, 10 blocked, and zero missing.

Batch 34 reviews the four remaining Azure and Windows Server freshness results. Twenty-seven first-party, distribution, upstream, book, expert, and community sources were registered and health checked. AZ-700 now distinguishes current virtual network flow logs from retiring NSG flow logs and labels agentless Connection troubleshoot preview. AZ-801 separates unaffected VM Insights performance monitoring from the retiring Map and Dependency Agent. AZ-800 and AZ-802 teach a cited provisional SSH Direct model across Debian/Ubuntu, Red Hat-family, SUSE/openSUSE, Oracle Linux, systemd, and libvirt while retaining the missing Microsoft end-to-end contract and compatibility matrix as a human-review concern. All four freshness results are current, leaving 212 current, zero review-required, 10 blocked, and zero missing.

Batch 35 revisits blocked CISM and Fortinet MSSP evidence under the best-effort rule. CISM's central November 2026 update remains unreadable, so an affiliated ISACA Ireland course is recorded only as a provisional architecture/strategy signal; NIST IR 8286 Rev. 1 and SP 800-61 Rev. 3 add transferable risk and incident practice. Fortinet's MSSP page still says Coming soon, so current FortiManager ADOM, FortiPortal multi-organization, and multi-agency CISA MSP/customer guidance strengthen the foundation without becoming inferred objectives. Both results remain blocked, leaving the baseline at 212 current, zero review-required, 10 blocked, and zero missing.

Batch 36 revisits all three MongoDB blocked results. Their public exam contracts and learning-path material remain useful, but each enrolled study-guide route still exposes only a generic client-rendered course shell to deterministic checks, so current detailed objectives cannot be claimed revalidated. The guides now label their September 2 weight or public-skill baselines precisely and require enrolled reconciliation. OWASP NoSQL security guidance, an older MongoDB data-modeling/performance article, and an AWS MongoDB Atlas landing-zone pattern add bounded practice without becoming inferred exam scope. All three results remain blocked, leaving the baseline at 212 current, zero review-required, 10 blocked, and zero missing.

Batch 37 revisits the three blocked Salesforce results and clears all three. Indexed official text now exposes the complete Agentforce Specialist and Platform Developer guides despite their simple-HTML Loading/CSS wrappers. Platform App Builder's blocker was a bad registered article ID: 005389157 now returns not found, while canonical 005298964 exposes the complete Summer '26 contract and five-domain objective list matching the committed snapshot. Salesforce architecture and certification-team material plus bounded ForceAcademy and Apex Hours cross-checks were registered and health checked; commercial or community material remains subordinate and no question content was used. The baseline is now 215 current, zero review-required, seven blocked, and zero missing.

Batch 38 revisits the two blocked ServiceNow results. Complete indexed text now exposes the January 2026 CSA blueprint, including its detailed six-domain scope and full exam contract, so that blocker is cleared. The CAD route still exposes only its title to deterministic and indexed extraction; its September 2 snapshot is therefore retained as an explicitly dated evidence baseline rather than presented as freshly revalidated. Current ServiceNow application-planning documentation, a ServiceNow employee-authored ACL article, and bounded community learning sources add useful practice without being used to manufacture exam scope. Conflicting secondary claims about weights, a fixed passing score, extra topics, or third-party mock tests are documented and excluded. The baseline is now 216 current, zero review-required, six blocked, and zero missing.

Batch 39 closes the remaining four-item candidate inbox. GH-600 is now a source-validated beta guide with its initial six-domain snapshot, twelve reachable official sources, seven safe or tabletop labs, and explicit beta/product volatility boundaries. GH-100 now uses the GHES release ledger for lifecycle checks and distinguishes Enterprise Live Migrations' bounded GHES-to-GHEC Data Residency path from normal GEI planning. Cisco's credential and exam pages still conflict on CCNA price and language; both are registered and cited, the narrower exam page remains the planning source, and checkout verification is required. The new GH-600 result is current, leaving 217 current, zero review-required, six blocked, and zero missing across 223 guides.