Day-of warm-up
The morning of your interview.
The night before through the morning after, in concrete actions. Designed to be printed (it's a single sheet) and stuck on your fridge or kept beside your laptop. Strip everything that's noise; keep what's actionable.
The night before
- Stop adding new material. Whatever you don't know tonight, you won't know tomorrow. Re-read your own notes for 30 minutes. Anything beyond that is anxiety, not preparation.
- Pack your gear. Laptop charger. Phone charger. Backup pen and paper. Water bottle. ID for the office (if in-person). Reading glasses.
- Pick your clothes. Comfortable, neat. Decisions in the morning are decisions you can't make about the round.
- Check the calendar. Time zones, video links, dial-in. Phone the interviewer's recruiter if anything is ambiguous; do it tonight, not in the morning.
- Set two alarms. One on your phone, one on a clock or a second device. Hotel alarm clocks are unreliable; trust nothing.
- Sleep by midnight. Caffeine clears in ~6 hours; alcohol degrades sleep quality even at one drink. Both are easier to forgo for one night than to regret.
Two hours before the first round
- Wake up early enough to be awake. Most people need 90 minutes between wake-up and clear thinking. If your first round is 10:00, be up by 8:30.
- Eat protein + slow carbs. Not just coffee. Eggs and toast, or oats with nuts, or yoghurt with granola. Sugar spikes followed by crashes are bad for working memory.
- Walk for fifteen minutes. Outside, ideally. Light + movement + fresh air resets cortisol. Do not check Slack or email.
- Review the trade-offs cheat sheet below. Nothing else. New material at this hour is anxiety, not study.
- Glass of water every 30 minutes. Mild dehydration costs ~5% on cognitive tests. You do not feel it; it shows up anyway.
Twenty minutes before
- Tool check. Camera works. Mic works. Speaker volume. Browser tabs open: your editor, the company's interview link, a blank Excalidraw or Whimsical board, your notes file.
- Close everything else. Slack, mail, music, notifications. The round needs the foreground.
- Bathroom break. The 45-minute round will feel like four hours if you skip this.
- Pen and paper at hand. Even if the round is on a whiteboard tool, having paper for capacity-math scratch saves time.
- Three deep breaths. Diaphragmatic, not chest. Slows heart rate, drops baseline cortisol. Costs nothing.
During the round
- Narrate before you draw. "Let me clarify the requirements first" — restate the question in your own words. Don't start drawing.
- Watch your clock. Phone or watch face-up beside you (interviewers don't mind). Cross-reference against the seven phases.
- Don't apologise for thinking. Ten seconds of silence to formulate is normal. "Let me think about that for a second" is fine; mumbled words to fill the void are not.
- Pick the deep dive proactively. Don't wait for the interviewer to ask. "Of these four components, this one is where the design lives or dies because X — I'll go deep here."
- If stuck, ask. "Am I going in the direction you expected?" gives the interviewer permission to redirect without it counting as a hint.
- Own your trade-offs. "I chose SQL here because the access pattern is mostly keyed lookups; the trade-off is harder horizontal scaling, which I'd address by sharding by tenant if we hit it." Cost named; mitigation named.
Between rounds
- Five minutes off. Water, bathroom, stand up. Look at the wall, not your screen.
- Do not replay the previous round. It's done. Replaying degrades your performance on the next one.
- Do not preview the next prompt. If you have the option to read prompts in advance, don't. It denies you the clarify phase, which is half the signal.
- Reset, don't recover. The goal isn't to feel better about the previous round; it's to enter the next one fresh.
After the loop
- Write down what you remember within an hour. Specifically: what you wish you had said, what you said well, what the interviewer pushed on. Notes you take an hour later are useful; notes you take a day later are mostly imagined.
- Don't message the recruiter the same day. They've heard from every other candidate already. Sleep on it. Send a brief, concrete thank-you the next morning.
- Don't replay the loop with friends. Use a journal or a single-page document for yourself. Talking the loop to death replaces real feedback with confabulation.
- Sleep early. The cognitive cost of a full loop is real. Take the next morning slow.
Trade-offs cheat sheet (one-pager)
The trade-offs you most need on the tip of your tongue. None of these is novel; the value is in having them already cached, so the round doesn't burn a minute deriving them.
| Knob | Cheap | Expensive |
|---|---|---|
| SQL vs NoSQL | Mature tooling, joins, transactions, ad-hoc queries. | Single-primary write ceiling; harder horizontal scaling. |
| Read-through cache | Cuts read load 10–100×; cheap operationally. | Stale data window; cache invalidation is a real problem. |
| Synchronous replication | Zero data loss on primary failure (RPO = 0). | Write latency = slowest replica's RTT. |
| Eventual consistency | Very high availability; horizontal scale. | Read-your-writes anomalies; conflict resolution complexity. |
| Sagas vs 2PC | No locks held cross-service; works at scale. | No isolation; compensations have to be coded for every step. |
| CDN / edge caching | p99 drops from hundreds of ms to ~10 ms for hot content. | Cache-key explosion if you vary on too many headers. |
| Microservices | Team autonomy; per-service scaling. | Network as a failure surface; distributed tracing operational cost. |
| Message queues | Failure isolation; smooths bursts; fan-out. | At-least-once delivery → consumers must be idempotent. |
| HTTP/3 (QUIC) | 1-RTT (0-RTT resume); no head-of-line blocking. | Middlebox compatibility issues; harder to debug than TCP. |
| Aggressive autoscaling | Pays only for what you use; absorbs spikes. | Cold-start latency; thrashing on noisy metrics. |
| Region failover (active-passive) | Cheaper; simpler. | RTO = minutes-to-hours; cold standby capacity sits idle. |
| Active-active multi-region | Near-zero RTO; bounded blast radius. | Conflict resolution; cross-region write latency; doubled cost. |
Numbers to know cold
| L1 cache | ~1 ns | (1 hour analogy) |
| L2 cache | ~4 ns | (4 hours) |
| Main memory | ~100 ns | (4 days) |
| SSD random read | ~100 μs | (11 years) |
| Disk seek | ~10 ms | (11 centuries) |
| Round-trip same DC | ~0.5 ms | |
| Round-trip cross-continent | ~80 ms | |
| Sequential SSD read 1 MB | ~0.25 ms | ~4 GB/s |
| Sequential disk read 1 MB | ~5 ms | ~200 MB/s |
| Network 1 MB over 1 Gbps | ~10 ms | |
| 2^10 / 2^20 / 2^30 / 2^40 | ~10^3 / 10^6 / 10^9 / 10^12 | |
| Seconds per day | ~86,400 | ~10^5 |
| Seconds per year | ~30M | ~3 × 10^7 |
See the full cheat sheets shelf for the printable A4 versions of these and other references.
Related on Semicolony
Found this useful?