Skip to main content

Cumulative Review (Semesters 0--8)

note

[TECHNIQUE 7: Interleaved Review] Mix all prior semesters (S0-S7) with Semester 8, weighted toward S8 design and leadership material. At least 15 of the 50 questions are drawn from S6 (data / distributed) and S7 (architecture / DDD) so long-term retention stays alive. Fifty questions; answer from memory first, then verify.

Questions​

  1. [S8 M1] Given only the prompt "design a photo-sharing feed," write the first five minutes of your response: functional requirements, three non-functional targets with concrete numbers, and the two hardest parts ranked.
  2. [S6 M3] In a primary-follower database with asynchronous replication, a user writes and then immediately reads and sometimes misses their own write. Name the failure, two common mitigations, and the cost each mitigation imposes.
  3. [S7 M3] What is a bounded context, and why does it matter more than the class diagram once a team grows past roughly 10 engineers?
  4. [S8 M4] A service has a mean response time of 120 ms and a P99 of 2,400 ms. Name three independent causes of a tall P99 tail and the metric that would confirm each cause.
  5. [S2] Why is a hash table's average-case O(1) lookup not a guarantee of O(1) latency in production? Name two real-world effects that break the guarantee.
  6. [S8 M2] When should you not extract a microservice, even when the boundary looks clean on paper? Give two concrete reasons.
  7. [S4] Roughly, what does a page fault cost versus an L3 cache miss? Why does that gap show up in system-design decisions?
  8. [S8 M3] State the difference between a choreographed saga and an orchestrated saga, and name one situation where you would pick each over the other.
  9. [S5] Why does a single TCP flow over a 50 ms round-trip path have a throughput ceiling even on a fast link? What raises it?
  10. [S7 M1] Write a six-part quality-attribute scenario (source, stimulus, artifact, environment, response, response measure) for "dashboard loads in under 2 seconds at peak."
  11. [S8 M5] Using Rumelt's structure, what distinguishes a strategy from a slogan? Give one sentence each for diagnosis, guiding policy, and coherent action in the context of "we should improve reliability."
  12. [S3] State the open-closed principle in your own words and give one concrete example where violating it created a measurable later cost.
  13. [S8 M1] Estimate the daily write storage for 100 million daily active users posting 5 times a day at 300 bytes per post. Show arithmetic and state one assumption you made.
  14. [S6 M4] Compare read-committed, repeatable-read, and snapshot isolation by naming one concrete anomaly each allows or prevents.
  15. [S8 M2] Two services want to share a database "for simplicity." What cost is being deferred, and what is the earliest observable signal that the deferral is coming due?
  16. [S1] Using birthday-paradox reasoning, roughly how many UUIDs do you need before collision probability hits 1%? Why does the answer matter for system design?
  17. [S8 M4] Write an SLI, SLO, and error-budget policy for "checkout completes in under 1 second at P95." State what traffic your team may reject when the budget is exhausted.
  18. [S5] Explain head-of-line blocking in HTTP/1.1. How does HTTP/2 change the picture, and where does HTTP/3 go further?
  19. [S8 M3] Name two guarantees a log-based broker (Kafka-style) gives you that a classical work queue does not, and one guarantee the work queue gives you that the log does not.
  20. [S7 M4] Describe a backward-compatible API change and a breaking one, with one example each, and state a rule of thumb for when to bump the major version.
  21. [S8 M1] List the stress tests you would run on your own high-level design before a reviewer does.
  22. [S6 M2] Why does an LSM-tree storage engine optimize write-heavy workloads, and what cost does it shift onto the read path?
  23. [S8 M5] Describe the four Staff+ archetypes (tech lead, architect, solver, right-hand) in one sentence each and name a signal that says "this is the archetype the org needs right now."
  24. [S0] Why do learners who write their own explanations outperform learners who only re-read? State the underlying learning-science reason in one sentence.
  25. [S8 M2] You publish an event named OrderCreated. Why does the name and payload shape matter on day one? Describe what breaks six months later if the contract is ambiguous.
  26. [S4] What does mmap give you that read/write does not, and when is it a trap rather than a gift?
  27. [S8 M4] Apply Little's Law: if a service handles 200 RPS and average latency is 50 ms, what is the average concurrency? Why does utilization above 80% behave non-linearly?
  28. [S3] Give a concrete example where favoring composition over inheritance prevented a design problem you would otherwise have hit later.
  29. [S8 M1] You have 10 minutes to pick a partitioning key for a URL-shortener's click-events table with 10 billion rows per month. What do you pick and why? Name one query pattern that choice makes painful.
  30. [S7 M5] What must an ADR contain to be worth writing, and what kind of ADR content is ceremonial noise?
  31. [S6 M5] Explain CAP under a network partition using one concrete user story -- not the CP/AP slogan -- that makes the tradeoff operationally real.
  32. [S8 M3] Describe the transactional inbox and outbox patterns. What specific failure does each prevent, and what operational cost do they add?
  33. [S2] When does quicksort's amortized analysis mislead you in practice, and what sort would you reach for instead, and why?
  34. [S8 M5] A VP says "we should just rewrite it." Describe your response in the meeting and in writing, staying honest about risks and avoiding defensive theater.
  35. [S5] What is the difference between a DNS TTL's intent and its real behavior across caches in the wild? Give one operational consequence.
  36. [S8 M4] Name three capacity-headroom strategies you could apply before a known 3x traffic event and the cost each one imposes.
  37. [S7 M2] Compare a modular monolith to a microservice split on four dimensions (deployment, data ownership, team coordination, failure isolation). When does each win?
  38. [S8 M1] On a whiteboard you draw a CDN in front of a service. What three questions should you be ready to answer about that placement before the reviewer asks?
  39. [S1] Why is Big-O dominance analysis insufficient when comparing two algorithms that both fit comfortably in cache? What do you measure instead?
  40. [S8 M2] Describe a backward-compatible evolution path for an API contract that currently returns userName as a single string and now needs to carry given name, family name, and display name.
  41. [S6 M1] Sketch (do not run) a SQL query that finds customers whose total spend last quarter exceeds the average spend of customers in their region, using a correlated subquery or window function.
  42. [S8 M5] What is the difference between mentorship and sponsorship, and why does a senior engineer need to practice both sides?
  43. [S4] What does strace show you that a profiler does not, and vice versa? Why does a production engineer need both in their toolkit?
  44. [S8 M3] Name event sourcing's two hardest operational costs and state what they buy you that a CRUD model cannot.
  45. [S7 M3] Give one example -- real or imagined -- where a team's ubiquitous language was wrong, and describe how the wrongness showed up in the code and in the bug tracker.
  46. [S8 M4] Describe how you would run a post-incident review that makes the team smarter rather than defensive. What do you include, and what do you deliberately leave out?
  47. [S8 M1] What belongs in the "open questions" section of a design doc, and what does its absence signal to a senior reviewer?
  48. [S2] A function is called one billion times with a 1% cache-miss rate, where a miss costs 100x a hit. What is the effective cost multiplier compared with the no-miss case? Why is this kind of arithmetic the core of performance engineering?
  49. [S8 M5] You have 30 minutes with a skeptical VP to justify a six-month platform investment. What is the structure of your memo, and what do you cut first if the meeting drops to 15 minutes?
  50. [Integration S6+S7+S8] For your RideHail Dispatch design: name the single technical decision in the design doc that will matter most in three years, the earliest observable signal that it was the right or wrong call, and what you would change about it now armed with Modules 4 and 5.

Answer Key​

Attempt every question closed-book first, then grade against your notes, the relevant module pages, and the primary texts. Record each miss as a one-sentence "what I got wrong and why" line and promote it to a spaced-repetition card the same day, tagged with the originating semester and module. Do not skip the closed-book attempt -- reading the question and then looking up the answer is comprehension practice, not retrieval practice, and the cumulative review only works as retrieval. Treat the review as passed when at least 40 of 50 are right on the first pass and the integration question (50) has a recommendation you would defend in a design review.