AIF-C01 deep review — September 28, 2026

Read the complete guide, 69 detailed objective bullets across 14 tasks, root/in-scope pages and revision history. The broad-domain snapshot matches the accepted baseline, but the guide misattributed newer material to revision 1.0. Correct it to revision 1.1, April 30, 2026. Italian/German delivery still retires after October 15; no new overall retirement was found.

Learner improvements

Add 35 answer explanations and five original worked cases: imbalanced classification, retrieval precision/recall, inference cost per accepted answer, subgroup results and train/evaluation leakage. Ten local arithmetic assertions passed; these build intuition without expanding the foundational target role into statistical implementation.

Explain Quick, Kiro, Strands and AgentCore through use cases. Add customization-data curation/governance and contrast Identity, Policy and Guardrails. Bind human approval to specific action parameters. Replace the shell-only model-evaluation URL with the working evaluation overview. Qualify A2I/Clarify new-customer closure on July 30 and Audit Manager new-account closure on April 30; Config does not reproduce every Audit Manager evidence/report capability.

The March 31 evaluation article is accepted for its introduction, evaluation modes and evaluator distinctions. The Strands introduction supplies framework vocabulary; its May 2025 roadmap is historical. Advanced code was not executed.

Remaining evidence limits

The revision history lists Clarify/Amazon Q additions that do not all appear in the corresponding live task/in-scope lists. Follow live objectives while recording this unresolved source discrepancy. No new blueprint version is inferred from it.

Kiro's direct extract was empty; a separately rendered public introduction supported its purpose. Pluralsight's November 2024, 2h4m course is a partial supplement. O'Reilly/Udemy were blocked; Whizlabs/Skill Builder yielded shells; video runtime/content was not verified. Tutorials Dojo lists 16 quizzes but markets actual-exam feedback: retain the explicit requirement to avoid recalled questions. MeasureUp's public description and automated-translation caveat were inspected; paid content was not.

Eight labs remain proposed. No AWS resources, live models, paid lessons or human evaluation were executed. Independent human review remains pending. The operational receipt records objective hashes, mappings, fetches and repository/site validation. Accepted snapshots were preserved.