System Design Camp — widening the scope without buying the hype
Built a /sdcamp page covering the full system-design spectrum across 6 layers, ordered by real importance. Started from a Gemini-provided 5-layer framing but corrected two things: added the design-method layer it omitted, and rejected its 'interview system design is only 20%' claim.
Context
The owner wanted system design widened beyond the interview slice (caching/sharding/queues) into a real training camp, and pasted a Gemini-authored 5-layer framing (code/process, application, infra [Hello Interview], agentic, business) asking if it was right.
What I rejected from the AI suggestion
Two things. First, the framing omitted the single most important layer: the method — how you actually drive a design (clarify requirements → estimate → define the API → data model → high-level architecture → deep-dive a bottleneck → articulate trade-offs). That's what an interviewer scores, and it was missing. Second, Gemini's rhetorical claim that "Hello-Interview-style distributed design is only ~20%, just FAANG exam questions" is overstated and, for someone with imminent Opendoor/Google interviews, actively misleading: the distributed/infra layer is not a footnote, it's exactly what gets you hired. I kept the spectrum idea (it is genuinely broader than the interview slice) but ordered by real importance and labeled the distributed layer as core, with an explicit honesty note on the page.
Chosen
A /sdcamp page with 6 layers, importance-ordered: (1) the method, (2) distributed building blocks [interview core], (3) data & application design, (4) reliability & business safety, (5) agentic system design, (6) process/resource micro. Each layer is a curated MAP over content that already exists — it pulls the relevant concept and AI cards by id and links the full /sd walkthroughs — rather than duplicating anything. Added 11 new verified concept cards for the genuine gaps, the biggest being the six method cards.
Trade-off accepted
The page is a heavier client bundle (it imports the concept/AI/SD data to render cards inline). Acceptable for a study tool. Kept all content universal — no personal war-stories fabricated, per the owner's instruction.
Verified by
- 11/11 new cards passed the fact-verify pass; CONCEPTS -> 295.
- tsc clean; next build compiled /sdcamp (243 static pages); every SDCAMP layer reference resolves to a real card id or /sd slug (checked at build).
- Deploy pending: SSH to the NCP box is currently blocked because the sandbox egress IP changed and the ACG isn't open to all; ships on the next firewall window with the frontend cards.
Commit
eadc86e8b0d31bac759229be18b1ad9754070dc9