Every drill is scored on the same 100–1000 scale as the real exam and measured against the 720 cut. So the question you keep asking — is one more practice test worth it? — finally has an answer.
20 questions · no card · full explanations either way
Twenty questions, allocated by the published domain weights so the drill has the same shape as the real exam. Free, no card, and you keep the explanations whether you buy or not.
After every sitting your mastery is re-estimated for each objective. The domains you're failing get more of the next drill — and they pull new items, not the ones you already missed.
Mastery times blueprint weight gives a scaled-score estimate on the real 100–1000 range, with a range that narrows as the evidence builds. When it clears 720 with room to spare, book the exam.
Six static mock exams, taken in order, scored as a percentage that means nothing on the real scale. That is the entire category. Here is what changes when the bank is driven by a measurement engine instead of a table of contents.
Fifteen objectives at equal weight, covering agent orchestration, subagent design and MCP integration. The questions test architecture decisions rather than API recall.
Seven domains for people using Claude at work rather than building on it. Output evaluation alone is a fifth of the exam.
Seven domains at production depth. Integration is the heaviest at 19%, and items test design judgement under competing constraints.
Eight domains, weighted to the official blueprint, so a 53-item drill has the same shape as the real exam.
A permanent free diagnostic, and a one-time price for the full bank. No renewal, because nobody renews a study tool after they have passed.