# BRAIN.md — Semantic Memory Layer

This file captures observed patterns, architectural decisions, invariants, and lessons learned from working within the invest-copilot codebase. It is updated iteratively as new structural insights emerge.

---

## Architecture Invariants

### Backend (FastAPI)
- **Entry:** `src/backend/main.py` — FastAPI app with middleware, CORS, lifespan events
- **Models:** `src/backend/models/` — SQLAlchemy ORM models (stock, price, watchlist, strategy, alert, sec_filing, insider_trade, screener, sector_rotation)
- **Schemas:** `src/backend/schemas/` — Pydantic v2 schemas for request/response validation
- **Routers:** `src/backend/routers/` — REST endpoints + `stream.py` for SSE/WebSocket
- **Services:** `src/backend/services/` — Business logic (auth, market_data, rotation, screener, sec, sentiment)
- **Tasks:** `src/backend/tasks/` — Celery workers (ingest_prices, ingest_sec, sector_scan)
- **Config:** `config.py` — Environment-based configuration
- **Database:** `database.py` — SQLAlchemy engine/session setup (PostgreSQL + TimescaleDB)
- **Cache:** `cache.py` — Redis cache layer
- **Storage:** `storage.py` — MinIO/S3-compatible object storage

### Frontend (Next.js 15)
- **App Router:** `src/frontend/src/app/` — Pages: dashboard, stock/[ticker], watchlists, strategies, screeners, sectors, alerts, sync, login, register
- **Components:** `src/frontend/src/components/` — UI primitives (shadcn/ui), layout (Sidebar, TopNav, Header), domain-specific (StockCard, PriceChart, StockTable, StockProfile, FilingCard)
- **State:** `src/frontend/src/store/` — Zustand stores (UI, Watchlist, Strategies)
- **Data Fetching:** `src/frontend/src/hooks/` — useStockData, useWatchlistData, useSSE
- **Lib:** `src/frontend/src/lib/` — API client, constants, utilities
- **Types:** `src/frontend/src/types/` + `src/shared/types.ts` — Shared TypeScript types

### Data Pipeline
- `src/data-pipeline/` — Python scripts for data ingestion (prices, SEC filings, financials, news)
- `pipeline.py` — Orchestrator
- `migration.sql` — Database schema migrations
- `tickers.json` — Ticker universe

### Infrastructure
- `docker-compose.dev.yml` / `docker-compose.prod.yml` — Docker Compose configurations
- `nginx/` — Reverse proxy configuration
- `.env.example` — Environment variable template

---

## Observed Patterns

### Data Flow (typical API request)
```
Frontend (TanStack Query) → FastAPI Router → Service Layer → SQLAlchemy Model → PostgreSQL
Frontend (Zustand) ← TanStack Query ← Router ← Service ← Model ← DB
```

### Real-time Data Flow
```
Celery Task (ingest) → Redis Pub/Sub → SSE/WebSocket (stream.py) → Frontend (useSSE hook)
```

### Auth Flow
```
Login → JWT token + session → Middleware validation → Protected routes
```

### Strategy → Alert Pipeline
```
User creates strategy (schemas/strategy) → Stored in DB → Celery worker evaluates → Alert created (models/alert) → Push/SSE notification
```

---

## Key Dependencies & Contracts

| Component | Depends On | Contract |
|-----------|-----------|----------|
| Routers | Services, Schemas | HTTP request/response via Pydantic |
| Services | Models, Cache, Storage | Business logic, data transformation |
| Models | Database | SQLAlchemy ORM, table relationships |
| Tasks | Services, Database | Async Celery jobs, idempotent ingestion |
| Frontend | Backend API | REST endpoints + SSE/WebSocket streams |
| Frontend Store | API hooks | Zustand state management |

---

## Security Considerations (Observed)
- JWT-based auth with session support
- CORS middleware configured
- Pydantic validation on all inputs
- Need to verify: rate limiting on API endpoints, input sanitization in screener queries, WebSocket connection limits

---

## Lessons Learned / Notes

<!-- Add observations here as they emerge from work sessions. -->
<!-- Example: "Price ingestion tasks must be idempotent — deduplicate by (ticker, timestamp) pair." -->
<!-- Example: "Zustand stores should mirror TanStack Query cache to avoid double-fetching." -->

---

## Pattern Attractors (Future Exploration)

- Sector rotation detection logic in `rotation_service.py` — how does it correlate with ETF flows?
- Screener query builder in `screener_service.py` — potential SQL injection surface if not parameterized
- SSE stream in `stream.py` — connection lifecycle and reconnection strategy
- Celery task deduplication — are ingestion tasks idempotent?