Claude Certified Architect — Professional glossary

12 terms the exam expects you to know, defined in plain language.

A

actionable error results(detailed failure messages, specific error details): Error results containing specific details that allow automated recovery decisions, unlike generic 'Operation failed' messages.

Detailed Error Response(detailed error messages): An error message that provides specific information enabling automated recovery decisions.
Deterministic Mechanism(deterministic, deterministic mechanisms): A programmatic enforcement that guarantees behavior, such as hooks, preconditions, schemas, or tool constraints.

Failure Categories in AI Agents(failure categories, failure categories in this domain): Distinct failure types including timeout vs empty result, syntax vs semantic errors, and drift vs overflow.
Failure Category(failure category, failure categories): Distinct types of errors such as timeout vs empty result, syntax vs semantic errors, and drift vs overflow.

Parallel subagents(parallel subagents): Subagents that execute concurrently from multiple Task calls in a single turn.
Probabilistic Mechanism(probabilistic, probabilistic mechanisms): A prompt-level encouragement that influences behavior but does not guarantee it, such as prompts or examples.
Probabilistic Mechanisms(probabilistic mechanisms): Mechanisms that encourage behavior but do not guarantee it, such as prompts and few-shot examples.
Programmatic Enforcement(programmatic enforcement): Using code-based controls to ensure behavior in scenarios with financial, legal, or safety stakes.
Prompt-Level Fix(prompt-level fix): Adjusting prompts or examples to solve an issue, suitable for non-critical tasks without financial, legal, or safety stakes.

Right-Sized Solution(right-sized solution, right-sized solutions): An appropriate level of complexity that avoids overengineering, such as using a description rewrite instead of a separate classifier model.
Right-Sized Solutions(right-sized solutions): The principle of choosing the simplest adequate fix rather than overengineered infrastructure for AI agent problems.