Matching Technique to Scenario-Based Tasks for the CCAR-P Exam

From heuristic to concrete tasks

The technique-selection heuristic is only useful if you can apply it to a real task in front of you. The Claude Certified Architect - Professional (CCAR-P) exam tests this at the apply level by handing you concrete tasks and asking which technique fits each. The skill is diagnosing where the difficulty of a task actually lives, in the format, in the judgment, or in the reasoning path, and choosing the technique that targets that specific difficulty, consistent with the lightest-that-works rule from technique selection by task complexity.

Four task shapes recur, and each has a natural technique. Learning the mapping, and the reasoning behind it, is what lets you answer these questions quickly and defend the choice.

The four canonical mappings

Fixed-category classification is zero-shot. A well-specified, low-ambiguity classification into a small fixed set of standard categories, such as routing a ticket into billing, technical, returns, account, or general, is work the model handles reliably from instructions alone. The categories can be described in words, so examples add cost without adding capability.

Variable-layout extraction is few-shot. Extracting structured fields, such as date, vendor, and amount, from inputs whose layout and formatting vary widely benefits from examples that demonstrate the target structure. Here the difficulty is that the output shape is hard to describe across so many input formats, and a few worked examples show it directly, which is exactly what few-shot is for. Balance those examples across the layouts the system will see, per bias introduced by prompt construction.

Multi-condition reasoning is chain-of-thought. A judgment that depends on several interacting conditions, such as whether a contract clause creates a liability under conditions that interact, benefits from chain-of-thought so the reasoning path is made explicit and checkable. The difficulty is the reasoning itself, not the output format, so the technique that targets reasoning is the right one.

Short summarization is zero-shot. Condensing a short passage into a couple of sentences is a simple, well-specified task the model handles reliably without scaffolding.

What the exam trips candidates on

The two traps are mismatches between the technique and where the difficulty lives. The first is applying chain-of-thought to a short, simple summarization task where zero-shot already produces a reliable, concise result. A scenario will over-scaffold an easy task; the credited reading uses zero-shot. The second is choosing few-shot for a multi-condition reasoning task when the harder problem is the reasoning path itself, not the output format. A scenario will address a reasoning problem with format examples; the credited reading recognises that the difficulty is the reasoning and reaches for chain-of-thought.

Common misreadings to avoid

How this shows up on the exam

Apply-level questions present one or more concrete tasks and ask which technique fits. The reliable answer diagnoses where each task's difficulty lives and matches the technique: zero-shot for fixed-category classification and short summarization, few-shot for variable-layout structured extraction, and chain-of-thought for multi-condition reasoning.

This knowledge point operationalises technique selection by task complexity using the prompt technique taxonomy, and it depends on bias introduced by prompt construction when the chosen technique is few-shot and the examples must be balanced.