Serverside Architecture Selector

Golden Principle: "The Last Responsible Moment"

• Simplicity is King: Always prefer the simplest solution that meets current requirements.
• Domain First: Do not make infrastructure decisions before defining domain requirements.
• Coupling vs. Abstraction: Favor simple coupling over premature abstraction until a pattern of repetition is proven.

1. Pre-flight Check

• Domain vs. Effort: If the domain complexity is High/Extreme, the agent MUST propose a draft of Bounded Contexts and require validation before suggesting any code patterns.
• Persistence and Performance: CQRS must not be suggested without a clear justification regarding "read/write mismatch".
• Legacy and ACL: If there is integration with legacy systems, an Anti-Corruption Layer (ACL) must be designed obligatorily.
• Microservices: Do not suggest microservices unless the cost of Team Cognitive Load or the need for independent scaling is the main bottleneck.

2. Architecture Definition (Decision Tree)

A: What is the complexity of your Business Domain?

• Low (MVP/CRUD): Transaction Script / MVC (Focus on simplicity and fast delivery).
• Medium (Clear rules, many integrations): Hexagonal Architecture (Side-effects isolation and testability).
• High (Complex domain, low change rate): Clean Architecture (Full framework abstraction and UI independence).
• Extreme (Complex, mutable, core-to-business domain): Domain-Driven Design (Ubiquitous Language, Bounded Contexts, and Living Business Logic).

B: What is the complexity of data access?

• Simple access (Simple CRUD, single database): Repository Pattern.
• Complex transactions (Complex transactions, multiple tables): Repository Pattern + Unit of Work.
• Read-Intensive / Audit (Complex domain, read-intensive, or audit trails): Data Mapper + CQRS (Command Query Responsibility Segregation).

C: What is the Communication and State strategy?

• Synchronous (Strong Coupling): Request-Response (Focus on immediate consistency).
• Asynchronous (Event-Driven): Pub/Sub or Message Queuing (Focus on decoupling and resilience).
• Distributed State / Auditing: Event Sourcing (The state is the sum of all past events).

D: What is the scale of the operation?

• Fast growth/Uncertain boundaries: Modular Monolith (Low infrastructure cost and easier refactoring).
• Massive scale/Independent teams: Microservices (Independência de deploy, poliglotismo, High scalability and tolerance for cost/complexity).

E: Who are the consumers?

• Public API / Multiple platforms: REST.
• Complex data needs / Multiple frontends: GraphQL.
• Real-time / Continuous data flow: WebSockets / Server-Sent Events.
• Internal microservices (Inter-service):
  • Low Latency / Performance: gRPC.
  • Simplicity / Universality: REST.

F: Evolution and Lifecycle Stage

• Initial Stage (Speed & Low overhead): Modular Monolith (Logical boundaries, shared database, simplified deployment).
• Growth Stage (Scaling specific modules or independent team growth): Extract Bounded Context to Microservice (Physical decoupling only when scaling needs differ).
• Optimization Stage (Highly granular, event-driven tasks): Serverless Functions / FaaS (Focused on side-effects and ephemeral workloads only).

G: Mobile & Multi-client Strategy (BFF Validation)

• Direct Integration: Use when the frontend is a single Web App and the API already returns optimized payloads.
• Backend For Frontend (BFF): MANDATORY if one or more of the following conditions are met:s
  1. Over-fetching/Latency: Mobile clients on 3G/4G receiving large JSONs where only 10% of fields are used.
  2. API Orchestration: The frontend needs to call 3+ microservices to render a single screen (Chatty I/O).
  3. Protocol Translation: Frontend needs REST/WebSockets, but downstream services use gRPC or SOAP.
  4. Client-Specific Logic: Mobile requires different data shapes, specific push notification logic, or image resizing not needed by Web.
• BFF Rejection Criteria (When NOT to use)
  1. Passthrough Proxy: If the BFF only forwards requests 1:1 to a single downstream service without transformation.
  2. Shared BFF: If you are building a "Universal BFF" for Web, iOS, and Android (this is just another API Gateway).
  3. Business Logic Leakage: If you find yourself implementing domain rules (discounts, taxes, auth-logic) in the BFF. These belong in the Domain/Microservice.

3. Review Checklist

• Does the architecture match the team size and cognitive load capacity?
• Is the BFF justified by specific UI needs or network constraints?
• Are there clear boundaries (Bounded Contexts) even in a Modular Monolith?
• Is the "Dependency Rule" pointing towards the Domain in all suggested layers?