v1.3.0

This release adds no new study material. Everything from v1.2.0 is still here: 320 practice questions, three mock exams per certification, 110 flashcards, the 33 page companion, and notes on all 21 official courses. What changed is that several things which looked correct were not, and the repository now proves them on every push rather than relying on care.

Corrections in the question bank

• Question ids were not unique. Each mock exam file numbers its questions from one, and the identifier did not record which mock it came from, so 180 questions across the four exams shared 60 ids between them. All 320 are now distinct.
• The three mock exams were indistinguishable. Every mock question was filed as mock, so the bank could not tell paper one from paper three. Each now names its own paper, and mock_exam.py gained --mock 1|2|3 and --practice, which makes the three papers separately sittable rather than pooled.
• Ten questions were filed under a scenario instead of a domain. Architect Foundations frames its practice inside the six published scenarios and labeled them "Scenario, Domain", but a comma also appears inside real domain names such as "Eval, testing, and debugging". The separator is now a colon, and the scenario is recorded in its own field.
• Two exams used domain names that appear nowhere in their blueprint. Architect Foundations filed questions under "Agentic Architecture" where the published domain is "Agentic Architecture & Orchestration", and Architect Professional under "Solution Design" where it is "Solution Design & Architecture". Per-domain scores were being split across buckets the real score report does not have: 15 values for a five domain blueprint, and 8 for a seven domain one. Both now match their blueprint exactly.

The build fails on a duplicate identifier, or on any domain that is not a name published on that exam's own study page, so none of these can return quietly.

The companion

The same 33 pages and 51 working links, rebuilt so the layout is deliberate rather than incidental.

• The content box now stops clear of the footer, which it had been running underneath.
• Illustrations scale to the height left on their page instead of a fixed cap, and the blueprint table shares its column, so an exam with five domains fills a page as evenly as one with eight.
• Part openers carry the pages in that part with real numbers, derived from the pages themselves rather than a list kept by hand.
• The cover is a mark and title lockup above a strip stating what is inside, each figure counted from the artifact it describes.
• Measured across all 33 pages: no page reaches the footer, and none ends short.

Corrections found by reading it end to end: page 19 and page 31 disagreed on a course title, because an ampersand was being rewritten as the word "and"; the glossary title-cased a domain the blueprint writes in sentence case; the closing page stated the same rule twice; Part 2's contents listed "Claude Certified Architect" twice with no way to tell Foundations from Professional; and four images carried no alt text.

Spelling

The repository mixed British and American forms, "behaviour" against "behavior" and "centre" against "center" among others. 86 occurrences were normalized to the American spellings already dominant, and a check now fails the build on nineteen watched forms.

Security

• The site no longer runs any third-party JavaScript. The star and fork widgets loaded a script that ran with full access to the page and, being a loader, could not carry a subresource integrity hash. They are now badges served as images from a host the page already used, so the counts are unchanged and nothing new is trusted.
• Third-party actions are pinned to commits rather than tags, because a tag can be moved to point at different code.
• The packages the workflows install are pinned exactly, and Dependabot watches those as well as the actions.
• update_resources.py refuses a source that is not HTTPS after redirects, and caps a response at 32 MB, since it writes into the repository from the network.
• Dependabot alerts and automated security fixes are enabled, and main is protected against force pushes and deletion, enforced on administrators.

Keeping the generated files honest

The question bank, tracker, and flashcards were already rebuilt and compared on every push. The images and the companion could not be, because they need a browser and the companion is set in a font the runner does not have, so a re-render would differ byte for byte with nothing actually wrong. Each render now records what it was built from, and the build fails if any of those sources changed afterwards. guide/maintenance.md documents all seven build commands, what each reads and writes, and the order to run them in when several change at once.

Verification

Eight checks run on every push. The site builds in strict mode with 51 pages and no dead internal reference, and all 195 external links resolve.

Nothing here is exam content. Everything is written from published material, and the official exam guides remain the authoritative source.

v1.2.0

Everything needed to prepare for all four Anthropic Claude certifications, in one place. This release adds the practice material, the website, and the printable booklet to what was previously a documentation repository.

Practice

• 320 original practice questions, 80 per certification, written against the published blueprints in the official item style: scenario-based, one best answer, and three distractors that each fail a stated constraint. Every question carries a rationale explaining why the other three are wrong.
• Three mock exams per certification, twelve in total. Each is 15 questions with its own answer key, domain breakdown, and readiness scale, and no question appears in more than one of them, so a second or third sitting measures readiness rather than recall.
• A practice engine in the browser and on the command line. It reorders questions and options on every run, so nothing can be memorised by position, and it scores per domain the way the real report does.
• 110 flashcards as a tab-separated file that Anki, Quizlet, and RemNote import directly, and as an interactive deck on the website filtered by exam or topic.

Reference

• A one-page cheat sheet per certification, with the blueprint weights and the rules that decide questions.
• A glossary of the 24 terms the exam guides use without defining.
• Per-course notes on all 21 official Anthropic Academy courses, with the order worth taking them in.
• Official videos, 16 of them arranged by the exam they serve, including Anthropic's complete AI Fluency course.
• Nine prompts for studying with Claude, each one click to copy: a weighted diagnostic, a domain drill, a post-mortem for questions you got wrong, a study planner, and one that reads a real score report and names the smallest gap to close.
• A progress tracker that weights each domain by its share of the exam, so the number reflects how much of the paper you can answer rather than how many boxes you have ticked.

The website and the booklet

• A documentation website at amey-thakur.github.io/CLAUDE-CERTIFICATIONS, 50 pages, searchable, with the practice engine, flashcards, and tracker running in the browser.
• A 33-page printable booklet as a single A4 PDF with 51 clickable links. It covers choosing an exam, the blueprint and cheat sheet for each, how to prepare, exam day, scoring and policies, the glossary, and the certificates behind it.
• Shareable images: a certification roadmap, four cheat sheets, six topic cards, and both faces of a flashcard, each carrying the repository link.

For Claude Code

• An exam coach skill and six namespaced slash commands: /cert:diagnostic, /cert:drill, /cert:mock, /cert:prep-plan, /cert:weekly-plan, and /cert:score-check. Clone the repository, open Claude Code inside it, and they quiz you directly from the blueprints.

Corrections worth naming

• Developer domain weights. The published blueprint lists eight domains totalling exactly 100.0%. The repository had merged two of them and rounded the rest, which summed to 101%. All four exams now carry the official figures and sum to 100.
• Course name. The Academy course is called Claude on Google Cloud, not Claude with Google Cloud's Vertex AI. Corrected everywhere it appeared.
• OnVUE system test. The bare system-test.onvue.com address is dead. The repository now uses the Anthropic-specific URL that Pearson's own program page links to.
• Answer distribution. The correct answer was B in almost every written question, which made the practice material gameable without reading the options. Answers are now spread evenly across all four positions.
• Certificate dates. All 21 course certificates now show their completion date, taken from each certificate's own verification record.

Verification

Every link in the repository, the website, and the booklet resolves: 1,044 source targets, 4,891 built-site references, and 207 distinct external URLs. Continuous integration lints the markdown, spell checks it, compiles every script, and regenerates the derived files to fail the build if they drift from their sources.

Nothing here is exam content. Everything is written from published material, and the official exam guides remain the authoritative source.

v1.1.0

• Documentation website at https://amey-thakur.github.io/CLAUDE-CERTIFICATIONS/, built with MkDocs Material and deployed from CI on every push
• Twenty-first certificate, AI Fluency for pK-12 Educators, with preview and verification link
• Certificate gallery with previews and Skilljar verification links for all courses
• Decision, troubleshooting, retake, and support-routing diagrams across the README and guide pages, with a readable palette in light and dark themes
• Citation metadata: CITATION.cff and codemeta.json
• Anthropic and Claude logo artwork from Wikimedia Commons, with attribution
• Root reorganized: community files, assets, scripts, and tool configuration moved under .github, leaving certification folders, the guide, and certificates at the top level
• All commits signed and following a uniform message convention; the dated record of changes now lives in these release notes

v1.0.0

Initial release.

• One folder per certification (Associate – Foundations, Developer – Foundations, Architect – Foundations, Architect – Professional), each containing the study guide with the full exam blueprint, the official exam guide PDF, and the maintainer's study notes
• Mirrored official PDFs: all four exam guides (version 1.0, July 2026), the Anthropic Certification Exam Policy, the Certification Terms and Conditions, and the Exam Registration Guide, with provenance recorded in guide/official-sources.md
• Learning paths page covering the CPN learning path, per-certification prep courses, the public course catalog, and partner badges
• Registration and scheduling guide, including OnVUE system setup, the network domain allowlist, and the application shutdown list
• Certification policies summary and condensed FAQ, including the June 30, 2026 Pearson and Credly migration notes
• Study strategy with per-exam emphasis and checklists
• Maintainer's Anthropic Academy course completion certificates with previews and verification links
• Maintenance guide, resource update script, issue templates, pull request template, and CI for markdown lint, spell check, and link checking