C07 Exercises¶
Three drills. Do them before reading ../solutions/ — a design you have
struggled with for 45 minutes teaches you something a design you have read does not.
| # | File | Clock | Skill |
|---|---|---|---|
| 1 | 01_design_video_moderation.md |
45 min | Apply the framework to a fresh prompt with a human-in-the-loop and an asymmetric fairness cost |
| 2 | 02_design_semantic_search.md |
45 min | Hybrid retrieval, reranking, freshness — the bridge between the two case studies and the base of any RAG system |
| 3 | 03_review_a_design.md |
30 min | Review a plausible but flawed design document and find five serious problems |
Do 3 last. Reviewing is a different muscle from designing, and it is closer to what a senior engineer actually spends the week doing.
How to run these¶
Set a timer and honour it. The constraint is the exercise. Producing a beautiful document in three hours trains the wrong skill; producing a defensible one in 45 minutes trains the right one.
Write it down, do not just think it. Use the ten-step framework from
../README.md §2 as literal headings. Filling every heading,
even with one line, is how you find the section you would otherwise have skipped.
Do the arithmetic on paper first, then check it. Both calculators accept the parameters these drills need:
The habit being trained is estimate first, verify second. If you reach for the tool before you have a rough number in your head, you are using it as a crutch rather than a check.
Score yourself against the rubric in
../README.md §9.4 before reading the solution, then
score again after. The gap between the two scores is the actual lesson.
What "done" looks like¶
A finished drill has all ten framework sections present, every quantitative claim traced to a labelled assumption, at least one latency budget whose hops sum to less than the SLO, and an explicit list of what you left out. It does not need to be right — the solutions deliberately point out places where a different choice would also have been defensible.
If you find yourself writing "this should be fast enough," stop and put a number there. That sentence is the single most reliable marker of a design that has not been thought through.