# Constraints

Search Engineering · Course

Interactive edition: https://mutapa.dev/search-engineering/course/constraints

Each chapter earns one constraint and hands it to the next, tracing how a request becomes an answer.

## Part 1: Search Foundations

### Chapter 1: [What Is Relevance?](https://mutapa.dev/search-engineering/course/what-is-relevance.md)

A system optimizes only for the signals that reach it. A metric can rise while the relevance it stands for falls.

### Chapter 2: [Corpus and Indexing](https://mutapa.dev/search-engineering/course/corpus-and-indexing.md)

No later stage can recover evidence the index discarded; indexing sets the ceiling for every stage that follows.

### Chapter 3: [Query Understanding](https://mutapa.dev/search-engineering/course/query-understanding.md)

A query is an incomplete description of an intent. Reading it well adds evidence about the request without changing the task it expresses.

### Chapter 4: [Retrieval](https://mutapa.dev/search-engineering/course/retrieval.md)

Because lexical and dense retrieval fail on different queries, a trustworthy candidate set needs both.

### Chapter 5: [Ranking and Fusion](https://mutapa.dev/search-engineering/course/ranking-and-fusion.md)

Ranking is the staged allocation of expensive attention across more candidates than any model can inspect, moving from cheap wide passes to costly narrow ones.

### Chapter 6: [Generation](https://mutapa.dev/search-engineering/course/generation.md)

Generation is bounded by its evidence rather than the model's fluency. Every claim traces to a source, or the answer becomes a refusal.

### Chapter 7: [Evaluation and Feedback](https://mutapa.dev/search-engineering/course/evaluation-and-feedback.md)

Behavioral feedback is biased evidence rather than ground truth, useful only when the full impression is logged and kept apart from interpretation.

## Part 2: Agentic Search

### Chapter 8: [Agentic Orchestration](https://mutapa.dev/search-engineering/course/agentic-orchestration.md)

An agent wraps the search system you already run instead of replacing it. Every retry pays for the quality of those tools.

### Chapter 9: [Plan and Route Complex Searches](https://mutapa.dev/search-engineering/course/plan-and-route-complex-searches.md)

A complex request is a graph of information dependencies, and progress comes from routing each part to a source that can answer it, then replanning from evidence rather than from motion.

### Chapter 10: [Maintain State and Recover](https://mutapa.dev/search-engineering/course/maintain-state-and-recover.md)

An agent recovers only because its verified progress lives outside the model as structured state; a failure then reruns only the unfinished work.

### Chapter 11: [Authorize Agent Actions](https://mutapa.dev/search-engineering/course/authorize-agent-actions.md)

Information, capability, and authority stay separate: retrieving a fact or holding a tool never grants permission to act, and authority comes from deterministic policy rather than model inference.

### Chapter 12: [Evaluate Agentic Search](https://mutapa.dev/search-engineering/course/evaluate-agentic-search.md)

A correct answer reached through a wrong path is still a failure. Evaluation weighs the whole trajectory and its cost, not only the result.

## Part 3: Governed Search

### Chapter 13: [Permissioned Retrieval](https://mutapa.dev/search-engineering/course/permissioned-retrieval.md)

Which evidence may this principal retrieve?

### Chapter 14: [Policy Through the Trajectory](https://mutapa.dev/search-engineering/course/policy-through-the-trajectory.md)

Authorization must shape candidate generation, not merely the final display. Once restricted evidence reaches ranking, caches, counts, or model context, hiding the result cannot undo the disclosure.

### Chapter 15: [Connectors and Trust Boundaries](https://mutapa.dev/search-engineering/course/connectors-and-trust-boundaries.md)

Authority is local to each protected decision. Every cache, memory store, delegated service, and resumed checkpoint must verify the current principal, audience, scope, policy version, and validity before using evidence.