06-reference/research

claude architect pro 2026 12 07 content delta

2026-11-02·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
anthropic-certificationccar-pexam-guidephdata-certstudy-sequencing

The post-12-07 CCAR-P content is not published, and there is no "v0.2" CCAR-P guide to diff against

The question

"What is the Claude Architect Professional exam's content actually changing to under the 2026-12-07 update (new Agent SDK / Skills / MCP coverage vs. the current v0.2 guide), ahead of the founder's 12-22 sitting?"

Context: the founder sits Claude Certified Architect – Professional (CCAR-P) on 2026-12-22, which is after the 2026-12-07 content change, with the updated guide expected 2026-11-17 — 15 days from today (date checked 2026-11-02 01:16 EST).

What we already know (from the vault)

What the web says

All of the following is from Anthropic's own Partner Academy, which is publicly readable without login (no partner-portal gate on these pages) — retrieved 2026-11-02 via Playwright.

Convergences and contradictions

Inference, NOT confirmed — the plausible-addition surface

Nothing below is in any published exam guide. It is derived from what Anthropic has shipped and started teaching since the v1.0 guide was written (July 2026), evidenced by the Academy catalogue as of 2026-11-02. Treat as a watch-list to validate against the 11-17 guide, never as a syllabus.

Academy courses that exist today but are absent from the CCAR-P prep course's prereq list, i.e. post-v1.0 surface: Introduction to agent skills · Introduction to subagents · Model Context Protocol: Advanced Topics (sampling, notifications, filesystem access, transports) · Introduction to Claude Cowork · Upgrade a client's build to the Claude 5.5 models · Working with Claude Fable 5.1 · Deploying Claude Enterprise with Confidence: the five decisions. Ranked by my guess at exam-blueprint impact:

  1. Agent Skills promoted out of D2's parenthesis. Today Skills is a sub-clause of "prompt reuse." A dedicated Academy course now exists. Most likely single change: Skills becomes its own objective, probably under D1 or D7.
  2. Claude Agent SDK appears at all. Zero mentions today and no Agent SDK course in the Academy catalogue ("SDK" = 0 hits on the all-courses page) — so this may not land on 12-07 despite being in the question's framing. If it does, D3 Integration (19%) is where it goes.
  3. MCP depth beyond protocol selection. D3 currently only asks you to choose MCP vs API/CLI vs agent-to-agent. The Advanced Topics course is the obvious feeder for a deeper objective.
  4. Model lineup refresh. D2 "select appropriate Claude models based on trade-offs" is version-coupled; Opus 5.5 / Fable 5.1 partner courses date the current wording.
  5. Enterprise rollout / governance (the "five decisions" course) mapping into D5 or D6, and Cowork as a non-code entry point in D1's "entry point" decision — both weaker signals.

Synthesis for RDCO

The honest answer to the question as asked is: nobody can know yet, and the brief's job is to stop him from studying the wrong thing for the next 15 days. The post-12-07 blueprint does not exist in public form, Anthropic has published two dates and no substance, and the "v0.2 guide" the question wants diffed is a different exam's document. What is now verified — and was not in the vault before today — is the complete pre-change baseline: the official v1.0 CCAR-P guide, its 7 domains and weights, its 63-item mechanics, and mechanical proof that Agent SDK coverage is currently zero while Skills and MCP are one objective apiece. That baseline is the thing the 11-17 guide gets diffed against, and it is worth archiving the PDF locally now because the S3 link is unversioned and will likely be overwritten in place on 11-17, destroying the only diff target.

Study sequencing: roughly 85% of this exam is safely studiable today, and it is not the technical 85%. Look at the weights: Solution Design 17 + Evaluation/Testing 16 + Governance/Safety/Risk 14 + Stakeholder Communication & Lifecycle 14 = 61% of the exam in domains whose objectives are almost entirely version-independent — RAG failure diagnosis, eval-as-acceptance-criteria, A/B testing, HIPAA/GDPR/FedRAMP mapping, human-in-the-loop routing, structured discovery, SLA conversations, handoff documentation. Add D3 Integration's non-surface-specific objectives (least privilege and capability bloat, auth/authz gap analysis, accuracy-latency trade-offs, chunking and indexing, observability at scale) and he can work most of the 19% too. This is also the part that plays against his known weak axis: his CCDV-F score report showed 0% on Agent Construction and MCP Development, but CCAR-P is a judgment-and-communication exam far more than an API-surface exam — the single D7 Claude Code mention is 7% of the paper. The 733-minute free prep course is the right spine for all of this, and the Stakeholder/GTM lesson (178 min, 14% of the exam) is the one he is least likely to already know from daily work. Defer only: D2's model-selection trade-offs (version-coupled), anything Skills-shaped beyond "modular prompts and caching," and MCP beyond protocol choice. That is maybe 2-3 hours of material. Do not pre-study an imagined Agent SDK objective — there is no evidence Anthropic even teaches one.

On sitting 12-22 versus squeezing in before 12-07: keep 12-22, and the strongest reason is one he has not been told. Guide §14 says that if exam content changes significantly, Anthropic may force holders to retake the full exam at full fee to recertify instead of taking the free non-proctored renewal. Passing the outgoing version ~two weeks before a known content change is precisely the scenario that clause describes, and it would put his December-2027 renewal at risk of a second $175 sitting. Sitting after the change buys a credential aligned to the current content for its full 12 months. The usual counter-argument — "sit early on content you can practice against" — does not apply here, because there is no CCAR-P practice exam either way; the readiness gate he used for CCDV-F (two fresh practice exams at 85%+) is unavailable, so sitting early buys him no extra calibration. A pre-12-07 sitting would also collide with the ~12-04 AIP-C01 placeholder. On bookability: registration checkout is live, the fee is $175 before partner discount, registration stays valid 5 years, and Pearson VUE allows reschedule up to 24h out — so he can and should register now (it locks nothing about content, which is set by test date) and schedule later. Whether a specific pre-12-07 seat is still available is the one thing I cannot verify without his Pearson login; it is moot under this recommendation.

One date to actually reconsider: 12-22 leaves zero retake buffer inside 2026. Per §11, a first fail triggers a 14-day wait — from 12-22 that is 2027-01-05, which would miss goals KR 4.1 (three Claude certs in 2026). Sitting 12-15 instead gives the same 28 days with the new guide, keeps a 12-29/12-30 retake inside the calendar year, and still sits on post-change content. The cost is a tighter gap after the ~12-04 AIP-C01 slot. That is a real trade-off for his judgment, not Ray's to decide — but if KR 4.1 is load-bearing for the review cycle, 12-15 dominates 12-22. Either way the critical path is: register now → work the 61% stable core from the free prep course for the next 15 days → re-diff the moment the 11-17 guide lands (that re-diff should be a scheduled task, not a hope) → confirm in-portal whether a CCAR-P practice mode exists.

Why this is in the vault

This is the first verified copy of the official CCAR-P blueprint in the vault — [[cert-progress]]'s 2026-10-21 entry explicitly carried CCAR-P's domains, fee and format as "unverified" third-party claims, and this replaces them with primary-source numbers. It also corrects a live premise error (the "v0.2 guide" is Architect–Foundations, not Professional) before that error propagates into a study plan, and it supplies the pre-change baseline that the 2026-11-17 guide must be diffed against for the 12-22 sitting on the main-bet cert escalator.

Open follow-ups

Related

Sources

Primary (Anthropic, retrieved 2026-11-02, publicly readable without login)

Vault

Third-party (consulted, NOT relied on for any claim above; listed for traceability)