SDSystem Design Studio
Search guide and handbook titles, headings, and text
Complete book · Compact glossary

Your handbook progress

0 of 31 sections complete.

Loading saved progress…

Guided learning paths

Interview preparation

Practice a repeatable design flow and the trade-offs most often explored in interviews.

12 sections · 3–5 hours · 0/12 complete

Continue path
  1. 1. Practical system-design workflow· Not complete
  2. 2. The 12-question system design loop· Not complete
  3. 3. 1. Requirements: FRs, NFRs, Constraints, and Assumptions· Not complete
  4. 4. 2B. Data Modeling, Indexing, and Partitioning· Not complete
  5. 5. 3. Concurrency· Not complete
  6. 6. 4. Transactions and Consistency· Not complete
  7. 7. 5. APIs, Contracts, and Idempotency· Not complete
  8. 8. 6. Messaging and Asynchronous Work· Not complete
  9. 9. 7. Failure Handling and Resilience· Not complete
  10. 10. 8. Scale, Capacity, Performance, and Caching· Not complete
  11. 11. 13. Master System Design Review Checklist· Not complete
  12. 12. Design review outcome template· Not complete

Architecture review

Review an architecture systematically from boundaries through operability and evolution.

17 sections · 5–7 hours · 0/17 complete

Continue path
  1. 1. 1. Requirements: FRs, NFRs, Constraints, and Assumptions· Not complete
  2. 2. 2. Boundaries, State, and Data· Not complete
  3. 3. 2A. Networking and Communication· Not complete
  4. 4. 2B. Data Modeling, Indexing, and Partitioning· Not complete
  5. 5. 2C. Time, Clocks, and Ordering· Not complete
  6. 6. 4. Transactions and Consistency· Not complete
  7. 7. 5. APIs, Contracts, and Idempotency· Not complete
  8. 8. 6. Messaging and Asynchronous Work· Not complete
  9. 9. 7. Failure Handling and Resilience· Not complete
  10. 10. 8. Scale, Capacity, Performance, and Caching· Not complete
  11. 11. 9. Security· Not complete
  12. 12. 10. Observability and Reliability· Not complete
  13. 13. 11. Deployment, Migration, and Evolution· Not complete
  14. 14. 12. Cost, Simplicity, and Operability· Not complete
  15. 15. 13. Master System Design Review Checklist· Not complete
  16. 16. Architecture Decision Record — short template· Not complete
  17. 17. Design review outcome template· Not complete

Agentic systems

Design agent and LLM systems with explicit contracts, failure boundaries, and review gates.

9 sections · 3–4 hours · 0/9 complete

Continue path
  1. 1. 1. Requirements: FRs, NFRs, Constraints, and Assumptions· Not complete
  2. 2. 5. APIs, Contracts, and Idempotency· Not complete
  3. 3. 6. Messaging and Asynchronous Work· Not complete
  4. 4. 7. Failure Handling and Resilience· Not complete
  5. 5. 9. Security· Not complete
  6. 6. 10. Observability and Reliability· Not complete
  7. 7. 15. LLM and Agentic Systems· Not complete
  8. 8. 16. Spec-Driven Development for Agentic Systems· Not complete
  9. 9. 17. Agent-System Design Review Checklist· Not complete

Complete handbook · Section 29 of 31

Compact glossary

Table · scroll horizontally when needed

TermSimple meaning
AvailabilityHow often the service is usable as required.
BackpressureA way to slow or reject incoming work when downstream capacity is exhausted.
Circuit breakerTemporarily stops calls to a repeatedly failing dependency.
ConcurrencyOperations overlap in time and may interact through shared state.
ConsistencyRules describing when different readers observe state changes.
DeadlockTwo or more operations wait on each other and cannot make progress.
Eventual consistencyDifferent copies or services may temporarily disagree but converge later.
IdempotencyRepeating the same operation has the same intended effect as performing it once.
IsolationHow strongly one database transaction is protected from effects of concurrent transactions.
LatencyTime required for one operation/request.
RPOMaximum acceptable data-loss window after a disaster.
RTOTarget time to restore the service after a disaster.
SLOTarget level of service reliability measured by one or more SLIs.
Source of truthAuthoritative owner/copy of a business fact.
ThroughputAmount of work completed per unit of time.
TraceTelemetry showing the path of a request across components.
TransactionA unit of related work committed or rolled back according to the data system's guarantees.
IndexAn extra data structure that speeds selected queries but costs storage and write work.
Partition / shardA subset of data managed separately to distribute load or capacity.
Partition keyValue used to decide which partition owns a data item.
SSEServer-Sent Events: an HTTP-based server-to-client event stream.
WebSocketA persistent two-way message channel between endpoints.
Background jobWork processed outside the original request, usually with explicit status and retry handling.
FR (Functional Requirement)What the system must do for a user, business process, or another system.
Acceptance criterionAn observable condition used to decide whether a requirement/story behavior is accepted.
ConstraintA boundary the design must respect, such as regulation, platform, budget, geography, or protocol.
AssumptionAn unproven statement temporarily treated as true and tracked for validation/review.
Clock skewDifference between clocks on different machines; it can make timestamps unsafe as a total ordering mechanism.
ReliabilityAbility to continue meeting required behavior over time and under defined conditions.
SLIA measured indicator of service behavior, such as availability or latency.
SLAA service-level agreement, usually a formal commitment between provider and consumer.
NFR (Nonfunctional Requirement)A measurable quality requirement for a named system or flow.
ADRA short record of an architecturally significant decision, its context, options, and trade-offs.
TIPThis book’s label for a Technical Implementation Plan; other teams may use “technical specification” or similar terms.
LLMLarge language model: a model trained to predict and generate language or other token sequences.
AgentA system in which a model can select steps and tools, observe results, and continue toward a bounded goal.
Agent loopRepeated model decision → action/tool → observation cycles until completion or a stopping condition.
WorkflowA mostly predefined control flow that may contain model steps.
OrchestrationCoordination of steps, agents, tools, state, budgets, and completion.
ToolA bounded capability exposed to a model through a machine-readable contract.
Structured outputModel output constrained to a declared machine-validated shape.
RAGRetrieval-augmented generation: generation supplied with relevant externally retrieved information.
EmbeddingA numeric representation used to compare or retrieve semantically related items.
RerankerA component that reorders retrieved candidates using a stronger relevance signal.
Context windowThe bounded input/output token space available to a model call.
Context compactionReplacing older or bulky context with a smaller representation that retains required state.
MemoryPersisted state reused across steps or runs; its type, scope, provenance, and retention must be explicit.
MCPModel Context Protocol: a protocol for connecting AI clients with tools, resources, and related capabilities.
A2AAgent2Agent Protocol: a protocol for discovery and task/artifact exchange between independently operated agents.
AGENTS.mdRepository-scoped Markdown guidance for coding agents.
Agent SkillA folder containing agent instructions and optional scripts, references, or assets loaded for a capability.
Evaluation / evalA repeatable task, environment, grader, and result used to measure system behavior.
GraderCode, a human rubric, or a calibrated model that scores an evaluation outcome or trajectory.
TrajectoryThe sequence of model decisions, tool calls, observations, handoffs, and state changes in a run.
GuardrailA control that checks or constrains input, output, tool use, policy, or runtime behavior.
Prompt injectionUntrusted content attempting to redirect a model or agent away from authorized instructions.
Excessive agencyGiving an agent more capability, permission, or autonomy than its task requires.
SandboxAn isolated execution boundary that limits filesystem, process, network, or other effects.
Human in the loopA designed pause where an authorized person reviews, approves, rejects, or redirects work.
Kill switchA fast control that disables a model, tool, agent, integration, or class of side effects.

Practice after learning

Section learning lab

Build the idea, test your recall, and keep page-specific notes.

The canvas is horizontally scrollable on narrow screens. Select a node and use arrow keys or the move controls; dragging also works.

Diagram ready.

Loading saved work…