Architecture Audit

Overview

Evaluate the structural health of a codebase — patterns, coupling, cohesion, and alignment between what the code claims to be and what it actually is.

Core principle: Architecture is not what's drawn on a whiteboard. It's what's in the code.

The Iron Law

code

NO ARCHITECTURAL CLAIMS WITHOUT READING THE ACTUAL CODE

When to Use

•"Audit this codebase"
•"Is this well-architected?"
•"What's the technical debt situation?"
•System feels hard to change or understand
•New team member needs to understand the system
•Before a major refactoring or migration

When NOT to Use

•First encounter with a codebase (use codebase-mapping first)
•Debugging a specific bug (use systematic-debugging)
•Performance bottleneck investigation only (use performance-audit)
•Need to understand what the codebase does (use codebase-mapping)

Anti-Shortcut Rules

code

YOU CANNOT:
- Assess architecture from directory names alone — read the actual code in each layer
- Claim "clean architecture" because folders are named well — check dependency direction
- Say "patterns are consistent" without checking EVERY module — inconsistency hides in corners
- Skip dependency graph analysis — coupling only reveals itself through imports
- Accept "it works" as architectural validation — working ≠ well-architected
- Audit only the happy path — error handling architecture matters equally
- Skip cross-cutting concerns — auth, logging, config are architecture too
- Trust documentation over code — code is the source of truth, docs drift

Common Rationalizations (Don't Accept These)

Rationalization	Reality
"The folder structure looks clean"	Folder structure ≠ architecture. Check dependency direction.
"We're using MVC/Clean Architecture"	Claiming a pattern ≠ implementing it. Verify with imports.
"Only the core modules matter"	Utility modules, middleware, and config can rot the architecture.
"It's a small project, architecture doesn't matter"	Small projects become large projects. Foundation matters.
"We'll fix the architecture later"	Later never comes. Architectural debt compounds exponentially.
"The tests prove the architecture is fine"	Tests can pass in a poorly architected system.

Iron Questions

code

1. Can I add a new feature without modifying existing code? (OCP)
2. Can I replace the database without touching business logic? (DIP)
3. Can I describe what each module does in ONE sentence? (SRP/cohesion)
4. Do inner layers depend on outer layers? (dependency direction violation)
5. If I change one module, how many other modules break? (coupling)
6. Are there circular dependencies? (sure sign of tangled architecture)
7. Does the code match the claimed architecture pattern?
8. Are cross-cutting concerns (auth, logging, errors) centralized or scattered?
9. Can I test business logic without infrastructure? (testability)
10. Is there a God class/module that everything depends on?

The Audit Process

Phase 1: Structure Mapping

code

1. MAP the directory structure — what's where
2. IDENTIFY the architectural pattern (layered, hexagonal, microservices, monolith, etc.)
3. COMPARE claimed architecture (docs) vs actual architecture (code)
4. DOCUMENT the real dependency graph

Key questions:

•Where does business logic live?
•Where does data access happen?
•Where are the boundaries between modules?
•What's shared vs isolated?

Phase 2: Dependency Analysis

code

1. MAP module dependencies — who imports what
2. IDENTIFY circular dependencies
3. CHECK dependency direction — do inner layers depend on outer?
4. MEASURE coupling — how many modules change when one module changes?

Dependency direction rules:

code

✅ Controllers → Services → Repositories → Models
❌ Models → Services (wrong direction)
❌ Services → Controllers (wrong direction)
❌ Utilities → Business Logic (wrong direction)

Dependency health matrix:

Metric	Healthy	Warning	Critical
Circular dependencies	0	1-2	3+
Max dependency depth	3-4 layers	5-6 layers	7+ layers
Fan-out (imports per module)	< 5	5-10	> 10
Fan-in (dependents per module)	< 10	10-20	> 20 (God module)

Phase 3: Pattern Consistency

code

1. IDENTIFY the dominant patterns (Repository, Service, Factory, etc.)
2. CHECK if patterns are applied consistently
3. FIND violations — where does the pattern break?
4. ASSESS — is the pattern appropriate for the problem?

Common pattern violations:

Pattern	Violation	Impact
MVC	Business logic in controllers	Untestable, duplicated logic
Repository	Direct SQL in services	Coupled to DB implementation
Service Layer	Fat controllers, thin services	Logic scattered across layers
Event-Driven	Synchronous calls bypassing events	Hidden behavior, tight coupling
Clean Architecture	Framework code in domain layer	Domain locked to framework
Microservices	Direct database calls between services	Shared state, no isolation
DDD	Anemic domain models (data-only entities)	Business logic leaks to services

Phase 4: Cohesion and Boundaries

code

1. CHECK module cohesion — does each module have a single, clear purpose?
2. CHECK boundary clarity — can you describe what a module does in one sentence?
3. FIND leaking abstractions — internal details exposed to consumers
4. IDENTIFY god classes/modules — things that do too much

Cohesion checklist:

Question	Healthy	Unhealthy
Can you describe this module in one sentence?	Yes	"It does X and Y and also Z"
How many reasons to change?	1-2	5+
How many other modules depend on it?	Few	Everyone
Can you replace it without changing consumers?	Yes	No
Are its internals hidden from consumers?	Yes	Consumers reach into internals

Phase 5: Cross-Cutting Concerns

code

1. HOW is authentication handled? (centralized or scattered?)
2. HOW is authorization enforced? (middleware, decorators, manual checks?)
3. HOW is error handling done? (global handler, per-endpoint, inconsistent?)
4. HOW is logging implemented? (structured, unstructured, missing?)
5. HOW is configuration managed? (env vars, hardcoded, mixed?)
6. HOW are transactions managed? (explicit, implicit, missing?)
7. HOW is validation done? (centralized, per-endpoint, duplicated?)

Cross-cutting concern health:

Concern	Centralized	Scattered	Missing
Auth	✅ Middleware	⚠️ Per-route checks	🔴 No auth
Errors	✅ Global handler	⚠️ Try-catch everywhere	🔴 Unhandled errors
Logging	✅ Structured, central	⚠️ console.log scattered	🔴 No logging
Config	✅ Environment + defaults	⚠️ Hardcoded in some places	🔴 Secrets in code
Validation	✅ Shared validators	⚠️ Inline checks	🔴 No validation

Phase 6: Evolution Assessment

code

1. HOW hard is it to add a new feature? (new endpoint, new entity, new UI page)
2. HOW hard is it to change an existing feature?
3. HOW hard is it to delete a feature?
4. WHAT breaks when you touch one thing? (blast radius)
5. HOW would this scale at 10x the current load?
6. WHAT would a new team member struggle with?

Output Format

markdown

# Architecture Audit: [Project Name]

## Executive Summary
[2-3 sentences: overall health assessment]

## Architecture Overview
- **Pattern:** [Identified pattern]
- **Claimed vs Actual:** [Alignment assessment]
- **Module Count:** [N modules/packages/services]
- **Primary Language(s):** [Languages]

## Dependency Map
[Key dependencies and their directions]

## Findings

### 🔴 Critical
[Findings with evidence and recommendations]

### 🟠 High
[...]

### 🟡 Medium
[...]

### 🟢 Low
[...]

## Metrics

| Metric | Value | Assessment |
|--------|-------|------------|
| Circular dependencies | N | 🔴/🟢 |
| God classes (>300 lines) | N | 🔴/🟢 |
| Dependency depth | N layers | 🟡/🟢 |
| Test coverage estimate | N% | 🟡/🟢 |
| Avg file length | N lines | 🟡/🟢 |
| Cross-cutting consistency | N/7 centralized | 🟡/🟢 |

## Recommendations
[Priority-ordered improvements with effort estimates]

## Verdict
[PASS / CONDITIONAL PASS / FAIL]

Red Flags — Stop and Escalate

•No clear boundaries between modules
•Circular dependencies everywhere
•Business logic in 10+ different layers
•No tests
•No consistent error handling
•Config values hardcoded throughout
•God module that everything imports from
•Domain layer imports framework/infrastructure code

Integration

•Before: codebase-mapping for initial understanding
•Alongside: security-audit, performance-audit, database-audit
•After: refactoring-safely for recommended improvements
•Triggers: writing-plans for remediation work