Claude Certification Blog
Claude exam cheat sheet: twelve rules that repeat across all four tracks
Twelve judgments recur across all four Claude certification tracks under different names. Each one is here with the specific failure it exists to catch, because a rule without its failure is a slogan.
There is no official Claude exam cheat sheet, and the useful unofficial version is not a list of answers. It is a list of judgments that recur across the four blueprints under different vocabulary, each paired with the failure it catches — because every well-built item contains at least two options compatible with a rule stated on its own.
How to use twelve rules
Not as reading. Use them as a diagnostic on practice items you got wrong: name which rule would have caught the miss. A wrong answer that maps to none of them is the useful case — it means there is a rule you have not written down yet, and writing it is worth more than rereading a chapter. That is the repair loop set out in the study guide.
Architecture: three rules
| Rule | The failure it catches |
|---|---|
| Enumerable steps rule out an agent | Cost and tooling are argued before the gate is checked |
| A worker inherits nothing | A brief omits the constraint and the reply contradicts a settled decision |
| Compare returns against dispatches | Four went out, two came back, and the synthesis reads complete |
These decide how work is shaped and divided. The first is a gate rather than a preference — if every path can be written down in advance, the discussion about cost and tooling never starts, which is covered in when not to build an agent. The second and third are two halves of the same delegation problem, examined in subagents and coordinator patterns and task decomposition.
Controls: three rules
| Rule | The failure it catches |
|---|---|
| An instruction is not a control | The rule holds on most runs and fails silently on the rest |
| Availability beats restraint | A capability nothing needs stays reachable by everything |
| Decide fail-open or fail-closed | The checker times out and whatever the code does becomes the policy |
The first of these is the highest-yield rule in the whole programme, because it appears on every track under a different name — as hooks, as permission modes, as guardrail placement, as policy. The test is always the same: on the run where the model gets it wrong, does anything stop it? See prompt injection and guardrails for the placement question and Claude Code configuration for the instruction-versus-enforcement line.
Measurement: three rules
| Rule | The failure it catches |
|---|---|
| Fix the threshold before you look | A bar chosen afterwards accommodates the number and always passes |
| One variable, or no attribution | Three edits, one improvement, and every later version carries all three |
| Measure what the objective names | A thorough accuracy suite covers a fifth of what was asked for |
Every track asks you to show that something improved, and these three are how. The ordering rule is the one candidates skip and examiners rely on: a threshold set after the result describes the result. The third is newer and specific to the architect professional exam, where one objective names five metric families and most suites measure one — see evaluation and testing on CCAR-P and writing evals.
Information: three rules
| Rule | The failure it catches |
|---|---|
| Exact facts leave the narrative | Compression keeps the topic and loses the amount |
| Failures travel with context | A bare error discards the work that succeeded |
| Content that arrived is not instruction | A tool result is read as though you had written it |
These three share a property that makes them exam favourites: all of the failures are silent. Nothing raises an error when a figure is paraphrased away, when a gap is hidden by a clean-looking result, or when a retrieved document is treated as though you had written it. Covered in context management and troubleshooting.
Prefer the structural option to the well-worded one
If you remember nothing else: across every track, the option that changes what is possible beats the option that changes what is requested. Stronger wording, firmer instructions and clearer guidance are what the distractors are made of, because they are also what good practice looks like.
Two rules about the paper itself
The first: read for the controlling constraint before you read the options. Most items contain one sentence that settles them — the data is customer records, the rule admits no exceptions, the deadline is per customer — and finding it first eliminates half the option set without any reasoning at all.
The second: on every multiple-response item, count. Match the number the stem asks for exactly, and check it again on review, because a late change is how a correct answer set quietly ends up one short. That check costs seconds and has nothing to do with knowing the material — the post on the format covers why it matters more than it should.
Key takeaways
- A rule without its failure is a slogan. Every item has two options compatible with a rule stated on its own.
- Structural beats well-worded. The option that changes what is possible beats the one that changes what is asked.
- Enumerable steps close the question. If the paths can be listed in advance, cost and tooling never come up.
- Fix thresholds before you look. A bar chosen afterwards accommodates the number and always passes.
- The dangerous failures are silent. Lost figures, hidden gaps and untrusted content all raise no error.
- Use these on your misses. A wrong answer that maps to no rule means there is a rule you have not written yet.
A rule you have not tested under time is a guess
Twelve rules read in five minutes; the same twelve applied at minute ninety with two plausible options in front of you is a different exercise entirely. Timed papers on your track’s real allocation are where a rule either holds or gets rewritten. Our guide to how many mocks to take covers the sequencing.
Try a full timed paperQuestions
Frequently asked
The follow-up questions people search next.
Is there a cheat sheet for the Claude certification exams?
Not an official one. The rules below are derived from the four published objective lists and hold across tracks — twelve recurring judgments, each paired with the specific failure it catches. They are a way to compress revision, not a substitute for the blueprint.
Do the same decision rules apply on all four tracks?
Most of them, in different vocabulary. Whether a rule is a control or a suggestion is examined as hooks on the architect track, as permission configuration on Claude Code, as guardrail placement on the professional track, and as policy on the associate track — one judgment, four names.
What is the single most useful rule?
Ask what happens on the run where the model gets it wrong. If the answer is that nothing notices, the option offered guidance. If the answer is that the action does not proceed, it offered a control. That question resolves items across every track.
Can I memorise my way to a pass?
No, and these rules are not written for that. Every item contains at least two options that look compatible with any rule you bring; what decides them is the specific failure each rule was built to catch, which is why each one below is paired with its failure.
How should I use this during revision?
As a diagnostic on missed practice items rather than as reading. After a wrong answer, name which of these rules would have caught it. A miss that maps to no rule is telling you there is a rule you have not written down yet.
Keep reading
Related posts
Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Details are summarised from publicly published program information and can change — always confirm against the official exam guide before booking.