Evidence and transparency

This appendix exists because the reverse question is fair. Five domain pages assert numbers, name traps and rank difficulties — and a reader booking a $125 exam deserves to know which assertions come from Anthropic and which come from us. The sorting rule is simple: a claim is confirmed if Anthropic prints it; cross-validated if we observed it in the sitting behind this guide and it is consistent with what Anthropic prints; ours if it is teaching structure or judgment. Nothing here is new material to learn — it is the metadata for the other five pages.

01Three evidence classes

02The official numbers — verified October 2026

We re-pulled the official exam guide on October 5, 2026 and checked every number these guides quote against it. The domain weights first, because the study plan follows from them:

The logistics, same check:

03The six official scenario contexts

Section 5 of the exam guide names all six scenarios the bank draws from — and, more useful, says which domains each one primarily tests. Your sitting sees four of the six at random, which is why two candidates can walk out with honest but opposite stories about what the exam emphasised. The six, condensed but faithful:

Now read the last column as a study map — this part is our inference, not Anthropic’s. Context management (D5, the lightest domain at 15%) is a primary domain in four of the six scenarios, more than any other domain — the random draw makes it unavoidable; nobody skips it by luck. The heaviest domain (D1, 27%) anchors three scenarios. Prompt engineering (D4) is primary in only two, but pairs with Claude Code configuration in the CI scenario, which is exactly where schema-enforced output meets automation. Study the domains by weight, then rehearse them through these six lenses: for each scenario you should be able to say what a good design looks like and the two ways it usually goes wrong.

04How wrong answers are written

The exam guide ships sample questions with full explanations, and the explanations do something most prep material does not: they classify every wrong option by its failure mode instead of just labelling it incorrect. Three families cover most of what those explanations show — and the same families recur in the traps sections of our five domain pages. The classification itself is cross-validated: consistent between the official explanations and the sitting behind this guide.

Practical use: when you eliminate options, name the failure mode out loud — “that is probabilistic enforcement”, “that solves availability, not ordering”, “that is over-engineering for a description fix”. An option you can classify is an option you can discard in seconds. The traps sections of the domain pages drill exactly this, one domain at a time; treat this table as the index.

05What is ours alone

• —The difficulty ranking — that CCAR-F is the hardest of the three certifications, ambiguity included — is one candidate’s experience across all three sittings. Anthropic makes no such claim.
• —The count “roughly 25 of the 60 questions are technical in wording” was tallied in a single sitting. One draw, one candidate: honest, but approximate by construction.
• —The study order (weight order, Domain 01 first) and the claim that hours should follow marks available — our judgment, argued on the Start here page.
• —Every trap in the domain pages: collected from the sitting and the official explanations, curated by us. Representative, not exhaustive — no independent guide’s trap list is exhaustive either, whatever the page count.
• —The glossary phrasing, the trees and flows, the quiz questions: written for this guide, not sourced from Anthropic.

06Verify us

Every confirmed claim on this page is checkable in minutes, and the checking is free. If you find a number here that contradicts the official guide, trust the guide — then tell us, because the error is ours.