Back to roadmap
Module 1 · Foundations & MethodDay 00625 min

Reading Existing Systems

How to reverse-engineer a deployed product into its design.

← PreviousNext →
Day 006

Reading Existing Systems

URL/headers
note
Latency
note
Feature behavior
note
Inferred design
service
Architecture model
datastore
Signal path
Inference path from public surface back to arch...
URL/headers
note
flow
Inferred design
service
Latency
note
flow
Inferred design
service
Feature behavior
note
flow
Inferred design
service
Memory hook

Reading Existing Systems: how to reverse-engineer a deployed product into its design

Mental model

frame the problem before drawing the system

Design lens

CDN delivers most bytes for ~$0.01/GB; recompute would cost orders of magnitude m...

Recall anchors
SurfaceTimingFeature signals

Why it matters

Designing systems gets easier when you can read other systems. From a product's URLs, response headers, latency, and feature behavior you can usually infer 70% of its architecture: where it caches, what shards by user, what runs at the edge.

Deep dive

Public clues. Cookie/auth headers reveal session vs token auth. CDN headers (cf-cache-status, x-amz-cf-id) reveal caching. URL shape reveals routing/sharding boundaries.

Latency clues. A 30ms response from Tokyo to a US-only product means edge cache. A 600ms first-byte every time means no cache or per-user dynamic content.

Feature clues. 'Likes' that update instantly for you but appear delayed to others reveal eventual consistency and probably a write-through cache. Rate-limit responses (429) hint at gateway sophistication.

Demo / scenario

Reverse-engineer a popular news site's homepage architecture.

  1. Static-ish article HTML → CDN-cached behind versioned URLs.
  2. Personal recommendations slot loaded async → separate API call to ranking service.
  3. Comments lazy-loaded → likely a separate microservice.
  4. Login uses JWT cookie → stateless app servers.

Tradeoffs

  • CDN delivers most bytes for ~$0.01/GB; recompute would cost orders of magnitude more.
  • Separate ranking service deploys independently but adds a network hop.
  • Lazy comments reduce TTFB but require client JS to render fully.

Diagram

URL/headers
Latency
Feature behavior
Inferred design
Architecture model
Inference path from public surface back to architecture.

Mind map

Check yourself

Loading quiz…

Sources & further reading

← PreviousNext →