Skip to content

MLA-C02 deep review — September 28, 2026

The complete guide and all 107 detailed skills across four C02 domains were reviewed. The broad-domain snapshot matches the accepted baseline. The review retains a lifecycle blocker: the September announcement plans January 14, 2027 GA, while the credential page still says TBD. September 29 beta delivery is a future scheduled event in this review.

Changes for learners

  • Add five original worked decisions covering authorized retrieval metrics, total compression cost, duplicate side effects, layered evaluation and compatible rollback; explain all 42 knowledge checks.
  • Replace vague API-key advice with the current distinction between session-bounded short-term tokens and long-term exploration credentials. Clarify that budget alerts are not spending caps.
  • Resolve CodeCommit's old onboarding uncertainty from its November 25, 2025 reopening entry.
  • Qualify new-account labs affected by Agents Classic and SageMaker feature maintenance. Keep C02's conceptual agent/evaluation/monitoring coverage.
  • Update provider observations: the public practice page now uses C02 domain labels, but still mixes 65 questions with the beta's 170-minute duration. No paid bank or complete course coverage was verified.

Practitioner articles

The August 21 RAG compression article supports a bounded cost/quality investigation. The guide adds synthetic arithmetic and explicit evidence validation, rather than adopting its benchmark as a promised result or treating temperature zero as a correctness guarantee. Its example code was not executed.

The March 31 AgentCore Evaluations article informed the distinction between controlled evaluation and sampled live observation. Reviewed sections cover evaluator scope, tool/answer outcomes and deployment considerations; the entire implementation and embedded media were not evaluated.

Evidence and limits

The operational receipt contains dated fetches, prior catalog entries, per-skill hashes and section references. Accepted objective baselines were preserved. A fresh candidate directory has no previous snapshots, so the monitor's initial changed flag is not itself evidence of upstream changes; explicit comparison found the broad objectives unchanged.

Sixteen local synthetic assertions checked expected arithmetic and contracts. No AWS lab, hosted model, paid lesson or linked sample deployment was executed. Human review remains pending. Repository, learning-catalog and strict-site validation are recorded after completion.