What You Need to Know
A retrieval strategy is how an assistant finds the information behind an answer. The choice rests on two axes. Data shape is the form the information takes. Query pattern is how people ask for it. Then ask which failure matters most: staleness, missed recall, wrong precision or latency. Pick the cheapest strategy whose failure mode is acceptable. Objective 3.5 designed one pipeline; this objective decides when it is the right tool.
Take a fencing contractor whose assistant reads a live stock system, policies, installation guides and code. One system routinely needs several strategies at once.
The Shape Axis
Data shape decides where truth lives.
Live structured data, such as databases, catalogues and account records, already answers precisely and currently through its own query interface. The right integration is a query tool against the system of record (text-to-SQL, an API tool). Embedding its rows produces a stale, structure-blind copy. It is wrong the moment it is indexed and has lost the structure that made queries precise. The contractor's stock count is fetched this way.
Narrative documents, such as policies, manuals and articles, have no query language. Semantic retrieval over well-chunked text fits them: the 3.5 pipeline's home ground. The contractor's repair policy belongs here.
Code and rigidly structured text, such as the contractor's quote calculator, sit between. Exact symbols and cross-references defeat pure similarity, so retrieval navigates structure: search, open, follow references.
The Query-Pattern Axis
The same corpus serves different questions differently:
| Query pattern | Cue in the stem | Native strategy | Fencing contractor example |
|---|---|---|---|
| Exact lookup | An identifier: part number, error code, clause reference | Keyword or filtered search against the right source | A gate hinge by part number |
| Conceptual | "Summarise", "explain", "what is our policy on" | Semantic retrieval over chunked text | "What is our policy on storm-damage repairs?" |
| Multi-hop / exploratory | Comparison across documents, unknown search path | Agentic retrieval: search, read, refine, search again | Post-depth advice compared across three suppliers' guides |
| Recall-critical | "Must not miss a relevant document" (compliance, legal) | Hybrid search plus reranking | A safety audit that must find every risk assessment |
Anthropic's context-engineering guidance frames the multi-hop case as just-in-time retrieval. The agent keeps lightweight identifiers and loads data as questions arise. It discovers what matters progressively instead of front-loading a guess.
When Agentic Retrieval Earns Its Price
Agentic retrieval spends latency and tokens to buy precision on questions no single query answers. The price is justified for exploratory and comparative work. It is unjustified for a lookup a metadata filter would answer in one hop.
The failure that matters decides. Recall-critical corpora justify hybrid search plus reranking, because a missed document is the expensive failure. Latency-budgeted lookups justify a single query, because a slow answer is theirs. The contractor's three-guide comparison earns the agentic loop. Its hinge lookup does not.
Hybrid Is a Design Answer, Not a Compromise
One system can combine a system-of-record tool for live lookups, semantic retrieval over manuals and agentic retrieval for comparisons. The fencing contractor runs all three. A small, stable, high-value corpus wants no retrieval at all: the cached prompt serves it whole.
The Selection Method
Work through four steps:
- Name the data shape.
- Name the query pattern.
- Name the failure that matters most.
- Pick the cheapest strategy whose failure mode is acceptable.
For the contractor's stock count: live structured data, exact lookup, staleness, so a query tool.
In the exam
A stem that presents three query families tests whether you match each to its native strategy or force one strategy onto all three. The exam keys mixed answers more often than pure ones. The selection method eliminates distractors mechanically. An option that embeds live operational data fails on staleness before any other property is examined. A prestige option that deploys agentic retrieval on a simple lookup fails on cost the question never justified. Domain 2 owns the cached-prompt boundary.
Exam Traps
Embedding live operational data — stock levels, balances, account records — into a vector store
Stale the moment it is indexed, and the structure that made queries precise is gone. The database already answers exactly and currently through a query tool; any option that embeds it is wrong on staleness alone.
Applying the retrieval approach that worked for the first corpus to every data shape that follows
Each shape has a native strategy. The transplanted one imports the wrong failure mode — stale rows, missed identifiers, weak recall — into a corpus that did not have it.
Deploying agentic multi-hop retrieval on a simple, latency-budgeted lookup
Latency and tokens spent buying precision the question never needed — the prestige distractor. Its inverse is also planted: single-query retrieval on a genuinely multi-hop comparison, which no one query can answer.
Choosing a precision- or latency-first option where the stem says no relevant document may be missed
Recall-critical language keys hybrid search plus reranking, because the missed document is the named expensive failure. Sacrificing the failure mode the stem names is the one unforgivable trade.
Practice Scenario
An enterprise assistant must answer three query families: the current stock level of part K-4402 (live ERP data), a summary of the data-retention policy (a 300-page manual), and how clause 14 has changed across the last three supplier contracts (an archive of several thousand contracts). Staff ask policy questions in everyday words that rarely match the manual's terms. Which retrieval design fits?
Build Exercise
Retrieval Strategy Matrix for One Estate
What you'll learn
- How to classify data sources by shape and queries by pattern
- How to name the failure that matters per source-pattern pair
- How to assign the cheapest strategy whose failure mode is acceptable
- How to defend an expensive assignment with a concrete breakage argument
- List the data sources one assistant must serve and classify each by shape: live structured, narrative documents, code or rigid text, small-and-stable
Why: Shape decides where truth lives, and the classification is the first elimination filter for candidate strategies.
You should see: A source list with one shape label each — and at least two different shapes, or the exercise needs a richer estate.
- For each source, list the real query patterns against it: exact lookup, conceptual, multi-hop, recall-critical
Why: The same corpus serves different questions differently — patterns, not sources, are what strategies serve.
You should see: A source-times-pattern grid with the cells that actually occur marked.
- Name the failure that matters most per source-pattern pair: staleness, missed recall, wrong precision, or latency
Why: The failure-that-matters is the tiebreaker between candidate strategies — it converts preference into argument.
You should see: One named failure per marked cell in the grid.
- Assign each pair the cheapest strategy whose failure mode is acceptable: system-of-record tool, semantic retrieval, hybrid plus reranking, agentic retrieval, or cached prompt with no retrieval
Why: This is the selection method the exam tests, run at estate scale: cheapest adequate, never prestige-driven.
You should see: A completed matrix, each cell holding a strategy and implicitly an argument.
- Defend the two most expensive assignments: what specifically breaks if the cheaper strategy is used, and how would the failure be detected
Why: An expensive strategy that cannot name its breakage scenario is prestige spending — and detection wires the choice into monitoring.
You should see: Two short defences, each naming the concrete failure and the observability signal (objective 3.4) that would catch it.
Sources
- Effective context engineering for AI agents — Anthropic
- Introducing Contextual Retrieval — Anthropic
- Claude Certified Architect Professional Exam Guide (v1.0) — Domain 3 — Anthropic