Cue-Sheet Template

For the interviewer. Use this when the candidate brings a prompt that is not in Question-Bank.md — a real question from a recruiter, a JD-flavoured problem, or something they invented. Building one takes about two minutes.

Write it into interview.md §10, not into a new file — it is scratch for the round. If the question turns out to be worth keeping, offer at debrief to promote it into the right archetype file as a Tier-2 compact sheet, and add a row to Question-Bank.md.

The shape 

### SD-nn · Title                                              [Tier 2 · archetype X]

**Read aloud:** *"One sentence. Deliberately under-specified. Names the user and the goal,
not the mechanism."*

**Frequency:** which companies ask this, and at what level.

**The crux.** One sentence: the single decision this problem is really about. Everything else is
scaffolding. *(This is the most important line on the sheet — it tells you where to spend deep-dive
probes.)*

**If they ask (give these, briefly and decisively):**
| Question | Answer |
|---|---|
| How many users / how much traffic? | *a number that makes the problem interesting* |
| Read- or write-heavy? | |
| Freshness / latency target? | |
| Retention? | |
| Payload size? | |
| Consistency needs? | |
| One region or global? | |
| Anything explicitly out of scope? | *give them one free cut, keep the rest for them to find* |

**A good design contains (listen for — not an ordering, not a completeness requirement):**
- 5–9 bullets in plain English. The mechanism, not the brand name.
- Mark **the senior insight** — the one observation that separates a strong answer from a
  competent one.

**Probe ladder:**
- *Requirements:* one probe that tests whether they cut scope deliberately.
- *Estimation:* one probe that forces a number into a decision.
- *Design:* the end-to-end walk, plus one contract question.
- *Deep dive (pick 2–3):* the crux · a failure · a scale jump · a hot-key/skew case ·
  a correctness-under-retry case.
- *EM bar:* how would you know in production · what does it cost · what would you cut ·
  how does this split across teams.

**Curveball:** one new requirement or failure to introduce ~65% through deep dives.

**Weak / Strong / Exceptional:**
- *Weak* — the naive design, named concretely.
- *Strong* — the expected competent answer.
- *Exceptional* — quantifies, sequences an MVP, owns the operational and cost story.

**Traps** — plausible-sounding wrong turns. Do not warn them; note it if they take one, and whether
they catch it themselves.

**Deep reference:** a public write-up, paper or guide section that covers this, if any.

Building it fast

  1. Find the crux first. Ask yourself: if the candidate got everything else right but this one thing wrong, would the design fail? That is the crux. Write it before anything else — the rest of the sheet follows from it.
  2. Pick numbers that make the problem interesting. The clarification answers exist to force a real decision. “A thousand users” makes every question trivial; “a billion” makes some questions silly. Choose the scale where the naive design just breaks.
  3. Write the weak band before the strong band. The naive design is easy to name and it is what you will actually be listening for the absence of.
  4. Put the mechanism, never the brand, in the checklist. “A durable partitioned log” — not “Kafka”. They should score for the mechanism whichever brand they name, and score nothing for a brand with no mechanism behind it.
  5. Six probes is plenty. You will use three or four.

Reusable probe bank

Independent of the problem. Draw from these when the sheet runs dry:

  • “Why that choice, and what did you trade away?”
  • “What are the two or three alternatives, and what’s the axis between them?”
  • “That component just died mid-write. Walk me through it.”
  • “Traffic 10×s next quarter. What breaks first?”
  • “One key is a thousand times hotter than the rest. What happens?”
  • “A client retries because it didn’t see your response. What does the system do?”
  • “How do you know, in production, that this is wrong — before a customer tells you?”
  • “What’s the SLO, and what’s the alert that fires before the SLO is breached?”
  • “What does this cost a month, and what’s the biggest line item?”
  • “You have half the time and half the team. What ships?”
  • “How would you split this across three teams, and what’s the contract between them?”
  • “What’s the one-way door here — the decision that’s expensive to reverse?”
  • “What would you have the team prototype first to de-risk this?”