Scoring Grid
How to convert a mock-exam raw score into a readiness judgement, and what to do about it.
1. What the real exam does with your answers
| Fact | Value |
|---|---|
| Items | 60 |
| Scale | 100–1,000 scaled |
| Pass | 720 |
| Scoring model | Criterion-referenced — a fixed standard from a standard-setting study |
| Reported | Pass/fail + scaled score + percent correct by domain |
| Domain percentages | Informational only — they do not gate the pass/fail decision |
Two consequences that matter for how you read these grids:
Criterion-referenced means you are not graded against other candidates. There is no curve, no quota, and no benefit or penalty from the cohort you sit with. The standard is fixed in advance. So a mock score is a meaningful estimate — it isn't contingent on who else is testing.
The raw→scaled mapping is not published. Anthropic does not release the conversion, and it can differ between forms because items differ in difficulty. So nobody — including this book — can tell you "44 raw = 720 scaled." What follows is a defensible estimate built on the standard assumption that a criterion-referenced professional certification sets its cut at roughly 70–75% of items correct. Treat the bands as guidance with real uncertainty, not arithmetic.
Practical upshot: aim comfortably above the estimated floor. The cost of over-preparing is a few hours; the cost of under-preparing is $125 and a 14-day wait.
2. Total-score bands
Based on Mock 1 (the blueprint-weighted paper).
| Raw / 60 | % | Band | Read |
|---|---|---|---|
| 52–60 | 87–100% | Safe | You are ready. Sit the exam. Further study has low marginal value — spend the remaining time on the flashcard decks and Cheatsheet 5, not on new material. |
| 48–51 | 80–85% | Comfortable | Ready, with margin for a bad form or a topic that lands awkwardly. Fix the specific misses and book the exam. |
| 44–47 | 73–78% | Estimated floor | Probably a pass, but without margin. One unlucky form and you're under. Do not book yet — close the domain gaps below first, re-sit Mock 1 after a week, and go when you're at 48+. |
| 38–43 | 63–72% | Not yet | A real gap, not bad luck. Identify the two weakest domains and re-read those chapters in full. Retest in 5–7 days. |
| < 38 | < 63% | Substantial work needed | Go back to the study plan and work the chapters in order. Don't take more mocks yet — you'll just re-measure the same gap. |
Target: 48+/60 before you book.
3. Per-domain minimums
Domain percentages are informational on the real score report — but a domain you're weak in still costs you raw items, and the raw items are what pass you. Use these as diagnostic thresholds, not as pass conditions.
| Domain | Weight | Items | Minimum | Comfortable |
|---|---|---|---|---|
| D1 — Agentic Architecture & Orchestration | 27% | 16 | 12 | 14 |
| D2 — Tool Design & MCP Integration | 18% | 11 | 8 | 9 |
| D3 — Claude Code Configuration & Workflows | 20% | 12 | 9 | 10 |
| D4 — Prompt Engineering & Structured Output | 20% | 12 | 9 | 10 |
| D5 — Context Management & Reliability | 15% | 9 | 7 | 8 |
| 60 | 45 | 51 |
Note that hitting every minimum sums to 45 — just above the estimated floor. That's the point: you cannot afford a domain that collapses. Two domains at 60% will sink an otherwise passing paper.
What each shortfall means
| Below minimum in | Most likely cause | Go to |
|---|---|---|
| D1 | The loop mechanics, or coordinator/subagent topology | Chapters 02–04; Cheatsheet 4; Deck 1 |
| D2 | Tool descriptions as the selection mechanism, or error taxonomy | Chapters 05–06; Cheatsheet 3; Deck 2 |
| D3 | Pure recall — file paths and flags. The cheapest domain to fix. | Chapters 07–08; Cheatsheet 2; Deck 3 |
| D4 | Explicit criteria vs exhortation, or schema-vs-semantics | Chapters 09–10; Deck 4 |
| D5 | Escalation triggers, or aggregate-metric traps | Chapters 11–12; Deck 5 |
If you're short on time, fix D3 first. It's 20% of the exam and almost entirely rote — paths, flags, frontmatter fields, and which files are version-controlled. It has the highest score-per-hour of anything on the paper.
4. Reading Mock 2
Mock 2 is not blueprint-weighted (D1 22 · D2 15 · D3 6 · D4 9 · D5 8) and its items are harder. Do not convert its total into a predicted exam score. Its job is different:
- Stress-test D1 + D2, which are 45% of the real exam
- Expose you to all six scenarios before exam day
- Give you harder discriminations than Mock 1 does
Expect Mock 2 to come in 3–6 points below your Mock 1 score. That gap is the paper, not you. Use Mock 2 for the D tags on your misses, and Mock 1 for the number.
5. Marking rules
- Multiple-response items are all-or-nothing. Three of four correct scores zero. The real exam states how many to select — if your count doesn't match the stem, you've already lost the item.
- Mark against the key before reading rationales. Get the number first, then learn from the misses. Reading rationales as you mark inflates your sense of how well you did.
- Score a guess you got right as a miss. If you couldn't have defended it, it isn't knowledge. Track these separately — they're the difference between a 47 that's really a 43 and a 47 that's really a 47.
- Time yourself. A 50/60 in 160 minutes is not a 50. The real constraint is 120 minutes for 60 items — 2.0 minutes each.
6. The miss log
For every item you got wrong or guessed, record one line:
Mock 1 #39 · D2 · Reflex 6 · access failure returned as empty result
Then group by the Reflex column. If four misses trace to Reflex 1 (deterministic enforcement beats prompt instructions), you don't have four problems — you have one, and it will cost you four items on the real exam. This is the single most valuable thing you can do with a marked paper.
The ten Core Reflexes are in Chapter 01 and tabulated in Cheatsheet 5.
7. Scheduling the mocks
From the study plan:
| Day | Action |
|---|---|
| 14 | Mock 1, full timed conditions. Mark, log misses, group by Reflex. |
| 15–19 | Work the gaps. Re-drill the decks for weak domains. |
| 20 | Mock 2. Read it as a diagnostic, not a prediction. |
| 21–22 | Final gaps + Cheatsheet 5 + the last-five-minutes list. |
| 23+ | Re-sit Mock 1 if you were under 48. Otherwise book the exam. |
Re-sitting Mock 1 after a week is genuinely informative — you won't remember 60 rationales, and the items that stick are the ones you actually learned. Re-sitting it after two days is not; you'll be recalling answers rather than reasoning.
8. Before you book
You are ready when all four hold:
- Mock 1 at 48+/60 under timed conditions
- Every domain at or above its minimum
- No Reflex accounting for 3+ misses
- You can state, from memory: the loop's termination condition · which config files are version-controlled · the three escalation triggers · what a JSON schema does and does not guarantee · the batch SLA arithmetic
Miss any of them and the remaining work is small and specific. Do it.