Right-Sizing
Assigning each pipeline stage the smallest model that clears its measured quality bar, rather than one model for the whole application. The decision gate is benchmark evidence on representative data, in both directions.
Exam context: Prestige selection (biggest model "to be safe") and blind downsizing (cost fix without re-running evals) are the paired planted wrong answers. Movement either way needs eval evidence.
See also: 2.1 Selecting Claude Models
Escalation Cascade
A routing design that defaults the request path to a small model and escalates to a larger tier on low confidence, failed validation, or explicit triggers. Because most requests never escalate, it buys near-big-model quality at near-small-model average cost. The name is this guide's shorthand; Anthropic's agent guidance documents the underlying pattern as routing (smaller models for easy requests, larger models for hard ones).
Exam context: The recurring keyed answer in cost-versus-quality stems, because it addresses both sides of the trade instead of sacrificing one. An expected escalation rate above roughly a third signals the small model does not clear the bar at all.
See also: 2.1 Selecting Claude Models
Effort Parameter
A control on recent Opus and Sonnet models that trades intelligence for latency and cost within a single model. Anthropic's model-choice guidance notes that tuning effort is often a better lever than switching models.
Exam context: The lever to check before reaching for a different tier. Distractors jump straight to a model swap when an in-model adjustment would meet the constraint.
See also: 2.1 Selecting Claude Models
Model Lifecycle
The published progression of Claude models through active, legacy, deprecated, and retired states, with at least 60 days' notice before retirement of publicly released models. Every model ID is a pinned snapshot; requests to retired models fail.
Exam context: The keyed process is pin the snapshot, monitor deprecation notices, and run an eval-gated migration. Letting an alias or provider default silently move the production model is the planted failure.
See also: 2.1 Selecting Claude Models
System Prompt Contract
The design stance that a system prompt carries everything true for every request — role, audience, standing policy, output contract — while per-request data lives in user turns. The split is doubly load-bearing: standing policy holds stable authority, and a stable prefix is what caching can reuse.
Exam context: Policy restated differently in each user turn drifts per developer, behaves less consistently, and pays full input price on every request. Moving it into the system prompt fixes both at once.
See also: 2.2 System Prompts and Guardrails
Escape Hatch
An explicit honest way out in the output contract: return null when the source lacks the field, or say the information is missing when the sources do not cover the question. Anthropic's guidance is that permission to admit uncertainty can drastically reduce false information.
Exam context: The first-line hallucination control. A schema that demands a value for every field forces the model to break the schema or invent — the fix is the null path, not a larger model or lower temperature.
See also: 2.2 System Prompts and Guardrails
Prompt Template
A prompt with named variable slots and XML-tagged sections separating instructions, context, examples, and per-request input. Turning a prompt into a template makes it a testable, versionable engineering artefact: diffable, code-reviewed, evaluated before deploy, rolled back on regression.
Exam context: One reviewed template serving a workflow beats ten drifting hand-copies. Tags also keep untrusted content arriving as data rather than instructions. Current state (checked 6 October 2026): for third-party content the prompt-injection docs now say to deliver it in tool_result blocks, never the system prompt or plain user text.
See also: 2.2 System Prompts and Guardrails
Layered Guardrails
Defence built in three layers around the model: instruction-level (scope, refusal rules, escape hatch), input-side (validation, sanitisation, delimiting untrusted content), and output-side (schema and business-rule checks before downstream consumption). The prompt is one layer, never the whole defence.
Exam context: Instructions shape behaviour probabilistically; only programmatic validation guarantees it. When a stem shows a system-prompt-only defence, the missing piece is code-level validation — and every refusal needs a fallback route.
See also: 2.2 System Prompts and Guardrails
Technique Ladder
The cost-ordered escalation of single-call techniques: clear zero-shot instructions first, few-shot examples second, chain-of-thought last. Each rung adds recurring tokens, so the cheapest technique that fixes the observed failure wins.
Exam context: Items pair a failure symptom with four plausible fixes; the tested skill is diagnosis. Examples for a reasoning failure, or reasoning for a format failure, spend tokens on the wrong problem.
See also: 2.3 Prompt Engineering Techniques
Few-Shot Prompting
Steering output with 3–5 worked examples wrapped in example tags. An effective set is relevant (mirrors the real use case), diverse (covers edge cases and near-misses), and structured (distinguishable from instructions).
Exam context: The canonical fix for format-compliance failures and classification boundaries. Five restatements of the happy path teach an unintended pattern; the set is only as good as its hardest example.
See also: 2.3 Prompt Engineering Techniques
Chain-of-Thought (CoT)
Instructing the model to reason step by step before answering, with tags separating the reasoning from the final answer so downstream parsing still works. Buys accuracy on multi-step reasoning at the cost of latency and output tokens.
Exam context: Reserved for failures that are genuinely about reasoning; on a high-volume trivial endpoint it is over-engineering. Visible reasoning is also an audit tool — it exposes faulty logic for review.
Current state (checked 6 October 2026): the prompting docs now say to rely on thinking on Claude Fable 5.1, Fable 5, Opus 5.5, Opus 5 and Sonnet 5.5, where a prompt asking for written-out reasoning (for example in <thinking> tags) may be declined. On the exam, prompted CoT is still the answer.
See also: 2.3 Prompt Engineering Techniques
Built-In Thinking
The reasoning capability inside current Claude models: adaptive thinking on 4.6-generation and later models decides when and how much to think, while the older Haiku 4.5 uses configured extended thinking. Thinking tokens bill as output and count towards max_tokens and the context window.
Exam context: Not a rival to prompted CoT. The architect's question is where reasoning happens and whether it must be auditable in the response body — prompted CoT keeps it there, though as of 6 October 2026 the docs say to rely on thinking instead on Claude Fable 5.1, Fable 5, Opus 5.5, Opus 5 and Sonnet 5.5.
See also: 2.3 Prompt Engineering Techniques
Context Rot
The documented degradation of accuracy and recall as token count grows, which makes curating what is in context as important as how much space is available. The goal is the smallest set of high-signal tokens that maximises the likelihood of the desired outcome.
Exam context: The reason "move to the bigger window and load more" is a planted wrong answer. Capacity solves fit; it amplifies rot, cost, and latency.
See also: 2.4 Context Window Optimisation
Replay Bill
The cost consequence of context accumulation: every token is billed on every request it appears in, so a 50k-token history replayed across twenty turns is a million billed input tokens. At volume, careless context becomes the dominant cost line, ahead of model choice.
Exam context: What makes context strategy an economics decision rather than housekeeping. Curation, caching, and compaction each attack a different part of the bill.
See also: 2.4 Context Window Optimisation
Compaction
Server-side summarisation (beta, 4.6-generation and later models) of earlier conversation turns into a compaction block once input tokens reach a trigger threshold, letting the conversation continue past the window limit. Anthropic's recommended primary strategy for long-running conversations and agentic sessions.
Exam context: The keyed answer for long-session stems where cost and late-conversation quality degrade together — it reduces context while preserving working state, where a fresh conversation destroys it.
See also: 2.4 Context Window Optimisation
Context Editing
A server-side mechanism (beta) that clears the oldest tool results, and optionally thinking blocks, once a threshold is reached, replacing them with placeholders. Reclaims window space occupied by stale content that adds cost and rot without adding information.
Exam context: The lever when tool results dominate the replay bill. Pairs with compaction in agentic-session stems.
See also: 2.4 Context Window Optimisation
Prefix Invalidation
The cache mechanic that prefixes build in the order tools → system → messages, and a change at any level invalidates that level and everything after it. The design corollary is front-loading stability: static content first, dynamic content last.
Exam context: A timestamp or request ID early in the prompt silently defeats the entire cache. Near-zero hit-rate stems are keyed to reordering content, not reconfiguring the cache.
See also: 2.5 Prompt Reuse Strategies
Cache Break-Even
The arithmetic that decides whether caching pays: writes cost 1.25× base input (5-minute TTL) or 2× (1-hour), reads cost 0.1×, so caching pays for itself after one read on the 5-minute duration and two on the 1-hour. TTLs refresh free on use.
Exam context: At volume the read discount dominates and caching cuts both cost and time-to-first-token; on sporadic traffic outside the TTL the write premium can exceed the savings. "Cache everything" is not universally correct.
See also: 2.5 Prompt Reuse Strategies
Modular Prompts
Composing prompts from versioned, owned sections (role, policy, task instructions, format contract, examples) assembled per feature from a shared library. One reviewed change to a shared section propagates everywhere; ten hand-copies drift apart and get fixed nine times.
Exam context: Also cache design: shared stable sections form common prefixes across features, and section ordering decides how much of the prefix caches. A module nobody owns drifts like a hand-copy.
See also: 2.5 Prompt Reuse Strategies
Agent Skill
A folder with a SKILL.md file: YAML frontmatter carrying a name (max 64 characters) and a description (max 1,024 characters) that doubles as the trigger, an instruction body, and optionally bundled files and scripts. Loads deep expertise on demand instead of paying for it on every request.
Exam context: The placement rule items test: standing policy belongs in the cached system prompt, on-demand expertise in Skills, per-request data in messages. Custom Skills do not sync across surfaces.
See also: 2.5 Prompt Reuse Strategies
Progressive Disclosure
The three-level token story of Skills: metadata only (~100 tokens per Skill) loads at startup, the SKILL.md body (under 5k tokens) loads only when triggered, and bundled resources cost nothing until read. Scripts execute with only their output entering context.
Exam context: The contrast is the monolithic system prompt that pays for every capability's instructions on every request whether used or not — the classic over-stuffing distractor.
See also: 2.5 Prompt Reuse Strategies