Anthropic's CCAR-F puts a scaled score on "can build agents"

Anthropic's credential program has a first exam with numbers attached: the Claude Certified Architect - Foundations, coded CCAR-F, launched in March 2026 ^[1][2]. The interesting part is not that a test exists, it is that a test forces someone to write down what "builds with Claude" means as weighted domains, which is the form a hiring bar or a training budget line can actually take ^[3][12].

The mechanics come from a dev.to explainer, which says the domain names and weights were checked against Anthropic's official exam guide ^[13]. Five domains: agentic architecture, tool and MCP design, Claude Code, prompt engineering, and context management ^[3]. Sixty questions in 120 minutes ^[5]. Pass mark of 720 on a scaled score with a 100 to 1000 range ^[4], which is 68.9 percent of the scorable band rather than the 72 percent the headline number suggests ^[1]. Items are multiple-choice and multiple-response, each stating how many answers to select, and they hang off four scenarios drawn from a bank of six rather than isolated recall questions ^[6]. That works out to two minutes per item ^[2] and roughly fifteen items per scenario ^[3], so the reading cost of each scenario is amortised across a block of questions.

The weights are the part worth planning against. D1, Agentic Architecture and Orchestration, is the single heaviest at 27 percent, with D3 and D4 tied at 20 percent ^[12]. Those three account for 67 percent, leaving 33 percent spread across the two domains whose weights the explainer does not name ^[4]. Despite the "Architect" title, the article describes the content as implementation-heavy and pitched at software engineers, solutions architects, and AI developers, validating that you can design and operate agentic systems on Claude rather than train models ^[7][8]. It is a Foundations exam, assuming general engineering maturity rather than years of LLM-specific work ^[15].

Two sources of confusion are worth clearing before anyone files an expense claim. The Pearson VUE registration listing carries four codes: CCAR-F for Architect - Foundations, CCAR-P for Architect - Professional, CCAO-F for Associate - Foundations, and CCDV-F for Developer - Foundations ^[9]. Third-party prep material tends to write the Architect - Foundations exam as CCA-F, or hedge with "CCA-F/CCAR-F"; per the explainer these are the same exam, and CCAR-F is the code Anthropic and Pearson VUE use, findable once you have Partner Network access ^[10][11]. Separately, this is not the AWS generative AI track: AIP-C01 and AIF-C01 test Bedrock, not Claude specifically ^[14].

Two gaps in the supplied material. Cost is listed as confirmed but no figure appears in the excerpt, so the price is not something I can report ^[16]. And the dates sit oddly together: a March 2026 launch ^[2], with the v1.0 exam guide described as published alongside a four-exam portfolio dated 2026-07-08, about four months later ^[5][5]. Either early sitters tested against an earlier document or the launch and guide dates are being reported loosely. Both come from the same single source.

What to watch: whether the other three codes actually open for registration, whether a published price and the two missing domain weights appear in the official guide, and whether CCAR-F starts turning up as a named line in job postings. A credential is only a hiring signal once someone outside the vendor writes it into a requisition.