> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lerian.studio/llms.txt
> Use this file to discover all available pages before exploring further.
# Architecture
> Explore Matcher's modular monolith built on DDD, hexagonal architecture, and CQRS, with eight bounded contexts that evolve independently.
Matcher is built as a **modular monolith** using Domain-Driven Design (DDD) and hexagonal architecture. CQRS separates commands (writes) from queries (reads).
This keeps operations simple while maintaining clear boundaries. Each module can evolve independently without the complexity of microservices.
## Architecture overview
***
## Bounded contexts
***
Matcher has six modules. Each owns its data and exposes clean interfaces to the others.
* **Configuration**: What you're reconciling (contexts, sources, field maps, rules)
* **Ingestion**: Getting data in (parsing, validation, normalization)
* **Matching**: The engine (rule execution, confidence scoring)
* **Exception**: Handling unmatched items (workflow, routing, resolution)
* **Governance**: Audit trails (immutable logs for compliance)
* **Reporting**: Visibility (reports, exports, dashboards)
### Configuration
Defines **what** you're reconciling and **how**.
**Handles:**
* Contexts (what's being reconciled)
* Sources (where data comes from)
* Field maps (translating external fields)
* Rules (how to match)
**Key models:**
* `ReconciliationContext`: The reconciliation scope
* `ReconciliationSource`: Source configuration
* `FieldMap`: Field translation rules
* `MatchRule`: Matching logic
### Ingestion
The Ingestion bounded context handles data intake and normalization.
**Responsibilities:**
* Parse uploaded files (CSV, JSON, XML)
* Validate incoming data against configured schemas
* Normalize external data into a canonical representation
* Detect and handle duplicate records
* Emit domain events when ingestion completes
**Key entities:**
* `IngestionJob`: Tracks ingestion lifecycle and status
* `RawImport`: Original uploaded payload
* `CanonicalTransaction`: Normalized transaction record
**Events published:**
* `IngestionCompleted`: Indicates data readiness for matching
### Matching
The Matching bounded context contains the reconciliation engine.
**Responsibilities:**
* Load applicable rules for a reconciliation context
* Execute matching strategies (exact, tolerance, date-based)
* Calculate confidence scores
* Create match groups and allocate transactions
* Identify unmatched transactions
**Key entities:**
* `MatchRun`: Execution of a matching job
* `MatchGroup`: Group of reconciled transactions
* `MatchItem`: Individual transaction allocation
**Events published:**
* `MatchConfirmed`: A match group has been finalized
* `TransactionUnmatched`: A transaction could not be reconciled
### Exception management
The Exception bounded context manages unresolved transactions.
**Responsibilities:**
* Classify exceptions by severity
* Route exceptions to internal teams or external systems
* Support manual overrides and adjustments
* Track resolution status and SLAs
* Integrate with external workflow tools
**Key entities:**
* `Exception`: An unresolved transaction
* `Resolution`: Outcome of exception handling
* `RoutingRule`: Routing and escalation logic
**Integrations:**
* JIRA for issue tracking
* Webhooks for custom workflows
### Governance
The Governance bounded context preserves reconciliation traceability.
**Responsibilities:**
* Record all system actions in immutable audit logs
* Provide queryable audit history
* Support regulatory and compliance reporting
**Key entities:**
* `AuditLog`: Append-only record of system activity
Audit logs are append-only by design. Entries cannot be modified or removed to preserve compliance integrity.
### Reporting
The Reporting bounded context provides operational visibility.
**Responsibilities:**
* Generate reconciliation reports
* Expose dashboard metrics
* Export reconciliation data in multiple formats
**Key entities:**
* `Report`: Reconciliation summary
* `Dashboard`: Aggregated operational metrics
* `ExportJob`: Asynchronous export execution
## Data flow
***
Reconciliation follows a deterministic pipeline across bounded contexts:
Reconciliation contexts, sources, field mappings, and rules are defined through the API.
External data is uploaded or fetched. Files are parsed, validated, normalized, and deduplicated. An `IngestionCompleted` event is emitted.
Matching rules are applied to eligible transactions, producing match groups with confidence scores. High-confidence matches are approved automatically. Unmatched items become exceptions.
Exceptions are classified, routed, and resolved either manually or via external systems. Resolution updates are propagated back to Matcher.
All actions across the pipeline are recorded in immutable audit logs.
Users access reports and dashboards showing reconciliation status, match rates, and exception aging.
## Infrastructure components
***
Matcher relies on the following infrastructure services:
| Component | Purpose | Usage |
| --------------- | ---------------------- | ------------------------------------------------------------------------ |
| **PostgreSQL** | Primary data store | Domain data with schema-per-tenant isolation |
| **Redis** | Cache and coordination | Deduplication, locks, idempotency keys |
| **RabbitMQ** | Message broker | Asynchronous communication between contexts |
| **Systemplane** | Runtime configuration | Hot-reload settings without restart via `/system/matcher/:key` admin API |
### Database architecture
* **Schema-per-tenant isolation** for strong data separation
* **Strong consistency** for matching and exception state
* **Eventual consistency** for reporting views
### Multi-tenancy
Matcher enforces strict tenant isolation:
* Tenant identity is extracted exclusively from JWT claims
* Tenant identifiers are never accepted via request parameters
* Database access is scoped through schema resolution
* All queries are automatically constrained to the active tenant
This model prevents cross-tenant data access and supports regulatory and audit requirements.
## Design patterns
***
### Hexagonal architecture
Each bounded context follows the ports-and-adapters pattern:
```
context/
├── adapters/
│ ├── http/
│ ├── postgres/
│ └── rabbitmq/
├── ports/
├── services/
│ ├── command/
│ └── query/
└── domain/
├── entities/
└── errors/
```
### Cqrs-light
Write and read paths are separated at the service level:
* `*_command.go` for state mutations
* `*_query.go` for read operations
This improves code organization and allows independent optimization of query paths.
### Outbox pattern
Event publication follows the outbox pattern:
1. Domain state and outbox records are persisted atomically
2. Background workers publish events to RabbitMQ
3. Events are marked as processed after successful delivery
This guarantees event delivery even during transient broker failures.
## Next steps
***
Explore the architecture through a guided example.
Review authentication, authorization, and tenant isolation mechanisms.