Architecture rounds,
made repeatable.

Level 1 — Simple

Client → API → DB

Correct for thousands of users. Start here out loud — it proves you scale on evidence, not habit.

Level 2 — Production

Client
  ↓
Load Balancer
  ↓
API (n instances)
  ↓
Cache → DB

The LB exists for deploys and instance failure; the cache exists because reads dominate. Each box has a stated reason.

Level 3 — Scale

Client
  ↓
CDN / LB
  ↓
Services (split by domain)
  ↓
Cache tier
  ↓
Partitioned DB

Partitioning appears only when one primary provably cannot hold the write volume. Services split along team and failure boundaries.

Level 4 — Global

Users
  ↓
Global routing (GeoDNS/Anycast)
  ↓
Regional services
  ↓
Regional caches
  ↓
Replicated data

Regions exist for latency and jurisdiction, not vanity. Now consistency between regions is your hardest problem — say so.