Skip to content
Exam guide · 6 min read

Exam Guide or Current Docs: Which Answer Does the Claude Exam Want?

The Claude exam guides date from July 2026 and the product has moved on. What Anthropic says about the guide, why exams lag, and which answer to pick.

AI Cert GuidesPublished
On this page7 sections

Every Claude certification has an exam guide, and all four current guides are version 1.0, effective July 2026. Claude Code, the API and the Agent SDK have all changed since. Sooner or later you will read something in the guide that the current docs describe differently, and you will wonder which version the exam wants.

Nobody has written a public answer to this, Anthropic included. But Anthropic's own certification pages say enough to settle it.

The problem, with one example

The Claude Certified Architect (Foundations) guide describes "the Task tool as the mechanism for spawning subagents", and says a coordinator's allowedTools must include it. Claude Code renamed that tool Agent in v2.1.63. Task still works as an alias, and the Agent SDK now emits Agent in its tool-use blocks.

That rename is harmless, because both names work. Where a guide describes a default, a limit or a list of options that has since changed, the guide and the current docs give different answers.

What Anthropic says

The guide defines the exam. The certification FAQ says the exam guide "lists the domains and task statements the exam covers and is the authoritative source for exam scope."

A question that doesn't match the guide is grounds for appeal. The certification policies list what an appeal can cover, including "a question you think was faulty or an exam that didn't match the published exam guide." If Anthropic confirms a question was faulty and it affected your result, "the remedy is a free retake rather than a changed score."

You can report a suspect question at no cost. From the same policies page: "If a question looks factually wrong, unclear, has more than one defensible answer, or doesn't match the exam guide, report it to Pearson." It adds that "reporting a problem never counts against you or affects your result."

The published guide is the standard the exam answers to. Anthropic doesn't promise the exam will match the docs, only the guide.

Why the exam lags the product

Each guide's header says: "This guide is subject to change without notice." When Anthropic describes exam changes, it describes them as new versions.

Each guide's credential maintenance section limits the credential to 12 months "because the underlying technology evolves rapidly". On-time renewal means you "review what has changed since you certified" and pass a free, non-proctored assessment. The policies page adds: "If we substantially update an exam, holders may need to pass the current version to stay certified." Retake waiting periods and attempt counts reset "when the exam moves to a new version."

So the exam is a snapshot taken at a known date, and renewal is how certified people catch up with what changed after it.

Other vendors build in a lag too. AWS reportedly holds a new feature off its certification exams until it has been generally available for six months, according to a Tutorials Dojo forum reply quoting the AWS Certification FAQ. We couldn't confirm that rule on AWS's own pages, but the reasoning is familiar: candidates need time to learn a feature before they're tested on it.

What to do on exam day

  1. Answer from the guide. When the guide and the current docs disagree, pick the answer the guide supports. That is what the exam is held to.
  2. Report anything that looks wrong. If a question seems to contradict the guide, or has two defensible answers, report it to Pearson. It costs you nothing.
  3. Appeal if it cost you a pass. Appeals go to Pearson support within 14 days of your exam date. The fix is a free retake, not a re-mark.
  4. Use current docs for real work. The guide tells you what the exam tests. It is not a manual for building things today. Code you ship should follow the live docs.

Don't assume every difference you spot is a trap. Most of the time the guide is simply silent on something newer, and the two sources don't conflict at all. Treat it as a conflict only when the guide states something the current docs contradict.

How we teach it on this site

We follow the same rule across all four certs: Architect (Foundations), Architect (Professional), Associate (Foundations) and Developer (Foundations).

  • Where a practice question's correct answer depends on a point the guide states, the answer follows the guide, even when the product has moved on.
  • Where the guide is behind the product, the lesson says so. You'll see a dated "Current state" note: what the guide says, what the product does as of the date we checked, a link to the doc, and which one to answer on the exam. The subagent invocation lesson has one for the Task to Agent rename.
  • Runnable code follows the current product. Code that only works against last year's API is no use to you, so build exercises use current model IDs and SDK calls.

When Anthropic publishes a new guide version, we compare it against the old one and remove the notes it makes unnecessary.

The short version

The exam guide decides what the exam tests. A question that doesn't match it can be reported or appealed, and exam changes come as new versions. On exam day, answer from the guide. Report anything that looks wrong. Everywhere else, use the current docs.

For what each guide covers, read our breakdowns for Architect (Foundations), Architect (Professional), Associate (Foundations) and Developer (Foundations).

Sources

Keep reading