MongoDB Associate Atlas Administrator deep review — September 29, 2026

The guide now answers 40 original prompts and includes 36 executed local checks, three worked scenarios and eight proposed Atlas activities. Thirteen public learning-path skill areas were reconfirmed; they are not detailed weighted exam objectives.

Scope and access

The main and course exam pages agree on 70 questions, 95 minutes, online proctoring, English, no prerequisite and USD 150. The course calls the questions multiple choice. The public path's two-hour exam card differs. No actual booking, passing percentage or policy interior was verified.

The canonical study-guide landing lists a free 30-minute guide and enrollment. The public Get the Guide viewer failed with 500 and its published asset returned 403. No document body, account access, enrollment or access-control alteration occurred. Historical scope/status snapshots remain intact, with hashes preserved. The automated monitor remains manual review. No dated replacement was identified in accessible certification material, which is not a full current objective audit.

Actual execution and boundaries

All 36 checks passed using standard-library Python and SQLite. A local SQLite database is backed up, receives a good update and a bad deletion, then is restored into isolation. The restored snapshot totals 300 cents and lacks the later valid 50-cent update. Explicit replay of that known fixture update recovers the intended 350-cent state without changing the damaged source. SQLite integrity passes. A separate counterexample shows that identical row count and sum can still contain the wrong identity. This is real SQLite execution, not Atlas backup, MongoDB oplog replay or a measured service recovery.

The access model separates network, identity, plane and action using trusted fixture values. It does not authenticate anyone. The inventory example preserves three identical usernames in different project/database scopes, rejects duplicate pages and detects missing records against a known local count. It is not a live API snapshot.

The original per-server connection exercise assumes a limit of 200, a target of 80%, 20 reserved connections and 12 processes. A pool cap of 11 yields 152 including reserve; 12 yields 164. No actual Atlas tier limit, driver topology or throughput is inferred.

Invented timestamps yield a 17-minute snapshot gap or a two-minute continuous recovery gap. Detection, decision, transfer, replay, validation and cutover sum to 47 minutes while transfer/replay alone is 25. The example exposes an incomplete recovery-time denominator; no Atlas RPO/RTO was measured. The alert model requires three regular above-threshold samples and treats missing/time-gap evidence separately. Those samples span two minutes, and no alert engine or notification was exercised.

No Atlas account, MongoDB server, CLI binary, API credential/token, network control, KMS, paid credit, infrastructure change or external message was used. The three CLI templates were reviewed against primary command syntax, not runtime-parsed or executed. Eight Atlas activities and independent human review remain pending.

Primary-source distinctions and discrepancies

The restore overview is Core-specific and requires preventing target client requests during restoration. Its CLI status section incorrectly gives atlas accessLists create for restore details and mislabels a detail command as a list operation. Dedicated list/describe references supply the guide's unexecuted templates. Their older singular examples and Project Owner requirement differ from their plural syntax headings and the overview's backup-specific viewing roles. Exact installed-version authorization and alias behavior remain unverified; no broad role was granted to resolve this.

The network guidance distinguishes API and database access lists. Its IP-authentication shorthand and API-key Terraform resource under project-access wording could blur that distinction. The guide requires separate identity, source and action evidence. Creating a private endpoint can leave public access enabled. The primary endpoint reference also announces non-port-mapped GCP endpoint retirement on April 30, 2027, a product watch rather than an exam version.

The KMS reference distinguishes network failure from invalid credentials or disabled/deleted keys. It includes edition-specific encryption/restore and search-rebuild limits. Broad recoverability or availability promises are not adopted. The current Atlas overview identifies Infinite as public preview; the familiar Core restore instructions must not be applied blindly across editions.

The backup guidance explains snapshot/oplog recovery and warming. Generic instant-access/minute-RPO language is not a measured result. The dedicated compliance-policy page documents protected retention, support verification, an empty-project exception and extra retention after the protected period. The alert reference gives a MongoDB 7.0+ dynamic-ticket caveat that takes precedence over its later generic low-ticket example.

The AWS landing-zone pattern's architecture is useful context, but its AdministratorAccess static-key instruction is not a least-privilege default. Its strong availability/encryption phrasing does not replace application tests or Atlas TLS requirements. No linked repository was cloned or deployed.

Selected CLI changelog entries identify 1.58.3 on September 10, 2026 and later command/API lifecycle changes. The first guessed changelog route returned 404; the corrected link was discovered from the CLI home page and returned current content. The old browser stable-route rendering looked stale, so fresh direct content and the current home selector supplied version evidence. Selected server 9.0 Upcoming headings remain a separate availability watch. Long release references were not comprehensively audited.

Catalog and evidence

Thirteen required path cards total 845 minutes, or 14h05, against a 13-hour headline. The separate 120-minute exam card makes 16h05, whereas both exam pages list 95 minutes. The practice landing lists six hours. Completion-discount advertising is recorded without account eligibility verification. The earlier v1 path's historical May 29 replacement note was not revalidated.

Three paid requests returned 403; earlier O'Reilly and Udemy durations remain unverified. Public cards are not proof of lesson quality or complete exam alignment. No practice-question interior was read or copied.

Thirty-one current direct receipts record 28 HTTP successes and three blocked requests. Several successes are shells; full browser reading and selected/reused primary-source boundaries are recorded separately. The failed public document attempt and earlier wrong changelog route remain private receipts. The operational record is ADLC_Docs/operations/2026-09-29-mongodb-associate-atlas-administrator-deep-review.json, preserving public-scope mapping, historical hashes, prior records, source boundaries, code/output and gates. The five reserved GitHub guides and disabled notification pilot remain untouched.