D28 Mock Interview Day: One Full China-Domestic-Style and One Full Overseas-Style Round, Self-Assessment

Today's Goals

Run through one complete China-domestic-style mock interview, from self-introduction to the candidate-questions segment
Run through one complete overseas-style mock interview, including a behavioral interview and system design
Score yourself with a rubric, and list at least 3 concrete weak points

Yesterday (D27) you packaged three repositories into resume bullets, READMEs, and a demo video. All of that is static: others can read it at leisure and you can revise it until satisfied, twenty times over with nobody knowing. Today is the opposite — sit down, start a timer, and from the first sentence there are no retakes.

Plain-Language Walkthrough

The domestic round: 63 minutes cut into five segments, each testing something different

Anybody who has taken a driving test has probably had the same experience: driving smoothly with the instructor, and stalling on the first move once you sit in the test car with an examiner holding a deduction sheet in the passenger seat. The car did not change and the route was one you practiced; the only difference is that somebody is now scoring you against a list.

An interview is exactly that scene. You explain your README fluently at your own computer, and with a stranger watching through a screen and writing in a notebook you start hearing your own voice waver by the thirtieth second. So rehearse before the real thing. But what actually improves you is not the rehearsal, it is that deduction sheet: it breaks an unpractisable verdict like "driving poorly" into a concrete action like "signaled for under 3 seconds before turning" that you can drill twenty times on its own. Today's whole output is your own deduction sheet.

First the shape of a domestic technical round. One round is usually a little over 60 minutes, cut into five segments, and in each the interviewer is judging something entirely different:

Segment	Duration	What the interviewer is judging	The most common deduction
Self-introduction	3 min	whether you can organize a structured piece of speech	starting from university and turning 3 minutes into reading the resume aloud
Project deep dive	25 min	whether you actually did these things	saying "we" throughout, so which part was yours is inaudible
Coding	20 min	whether you think aloud while writing	writing silently for 15 minutes and speaking only afterwards
Scenarios and fundamentals	10 min	whether your knowledge has a sense of boundary	bluffing rather than admitting ignorance, and collapsing two follow-ups in
Candidate questions	5 min	whether you genuinely thought about what you would come to do	asking one question answerable from the website

Of those five, project deep dive's 25 minutes weigh most, and it has one criterion: did you actually do this. The method is not listening to what you did but to whether you can say why you did not do it another way. Doing something leaves traces of choosing — you know two other roads existed, you know their costs, and you know why you dropped them. Somebody who did not do it can only restate the result and starts inventing at the second follow-up. So D27's STAR formula gets inverted here: the resume carries conclusions, and the interview asks about the process that produced them.

Coding's 20 minutes are also often misread. Whether you can write it obviously matters, and equally important is whether you make any sound while writing: every silent minute is a blank to the interviewer, and once you say your reasoning aloud they often pull you back when you head the wrong way — that is not cheating, that is them watching whether you can collaborate.

The time boxes are hard: the 2 extra minutes you spend on your self-introduction are deducted from the project deep dive, with no compensation. So "running out of time" is not an accident, it is the inevitable result of not rehearsing.

The trouble is that you feel none of this while inside it. Under stress a person's self-evaluation fails: afterwards all that remains is a vague "not bad" or "that was a disaster," and there is nothing to practice the next day.

The overseas process: five rounds, each with different people judging you from scratch

The overseas process's biggest difference from the domestic one is not the questions but the structure. Domestically it is commonly two or three technical rounds in one day with the same people digging deeper; overseas it is a string of independent stages spread over weeks, each owned by different people judging separately and voting at the end.

Round	Duration	What this round really examines	Common traps
recruiter screen	30 min	background fit, motivation, salary range, work authorization and time zone	treating it as small talk and inventing a salary range on the spot
technical / coding	60 min	thinking aloud, edge cases, complexity	silence throughout, discovering the approach was wrong in the last 5 minutes
system design	60 min	clarification, trade-offs, failure modes	the five steps reaching only the architecture diagram, with no time for trade-offs
behavioral	45 min	STAR, especially the number in the R	one story for three questions, none of them with an R
team match	varies	a two-way choice: you are interviewing this team too	assuming it is a formality and preparing no questions

The first round, the recruiter screen, is the most underestimated. It is run by a recruiter rather than an engineer and asks why you want to move, your expected salary range, and which time zone you can work in — and it is a real gate where a vague answer ends the process. The salary range especially deserves a range and a rationale prepared in advance, because a number invented on the spot weighs on every later stage.

Behavioral's 45 minutes are what domestic candidates find least familiar. It asks about experience rather than technology: tell me about a time you disagreed with a colleague, a time you got something wrong, a time you decided with incomplete information. The way to handle it is preparing 2 to 3 stories in advance and reusing them — one experience answers different questions from a different angle. Write the three, run each through D27's STAR formula, and check the R especially: an R with no number is no R. And guard against one trap in the other direction: do not answer three questions with one story, because the interviewer remembers.

The system-design round connects directly to D26's five steps and time boxes (5 minutes clarifying, 3 estimating, 8 sketching, 15 going deep, 5 on trade-offs). Overseas rounds care especially about those last 5 minutes: what you gave up, and at what scale this design gets overturned. A candidate who stops once the diagram is finished usually gets the verdict "can implement, judgment not visible."

The final team match is not a formality. It decides which group you join and what you work on, and in that round your questions carry more information than your answers: what problem this group finds most painful now, what the release cadence is, how code review works. Asking those reads as taking the job choice seriously; not asking reads as wanting any offer.

What both processes share: no round rewards improvisation. All of it consists of rehearsable actions.

How to score yourself so it is not a vague "fine"

Most people have no interview partner. The good news is that a mock interview's core value does not come from the person opposite, it comes from the replay afterwards, and a replay is something you can do alone.

The minimum kit for running one round alone is three things: a recording (audio or screen), a countdown timer, and a written question list. The order matters too — set alarms to the time boxes in the two tables above, write the question list in advance (self-introduction, one deep-dive question per project, one coding question, one system-design question, two behavioral questions), and then go from start to finish in one pass, with no pausing, no retakes, and no looking things up. One interruption voids that recording, because a real interview has no pause button.

To have somebody ask follow-ups, let a model be the interviewer. What matters is not elegant prompt wording but pinning down the follow-up discipline first, or it becomes an assistant that praises you endlessly:

TextText

Role: you are the interviewer running the project deep-dive segment for an agent
engineer role, 25 minutes total.
Discipline 1: ask one question at a time, then stop and wait for my answer; do not
answer your own questions.
Discipline 2: after each answer of mine, you must follow up one level deeper based on
what I just said, following up three levels in a row before changing topic.
Discipline 3: throughout, do not evaluate, do not praise, and do not fill in what I
left out.
Discipline 4: at the end of the segment, score me 1 to 5 on five dimensions, and point
out which sentence of mine triggered each deduction.
Material: below are my resume bullets and a one-line description of each of three projects.

Score after running it, and score against the replay rather than on how it felt at the end. Self-evaluation is most distorted in the ten minutes after finishing: a smooth run feels excellent and one stumble condemns everything. The rubric fixes five dimensions at 1 to 5 each, out of 25:

Dimension	1 point	3 points	5 points
Self-introduction	over time or off-topic, and afterwards nobody knows what you can do	finished within 3 minutes with structure, and the highlights rest on adjectives	the main line in 90 seconds, every highlight carrying a number, naturally inviting the follow-up they wanted
Project depth	says only what was done and cannot say why it was chosen	states the approach and trade-offs and survives two follow-ups	volunteers what was given up, the failure modes, and at what scale it gets overturned
Coding	long silences, or it does not run when finished	writes a working version and fills in edges when prompted	thinks aloud, volunteers complexity and three edge cases, and walks through it afterwards
System design	starts drawing boxes with no clarification	walks all five steps and stops at the diagram with no trade-offs	clarification with numbers, a grounded estimate, well-picked deep dives, and a close stating what was given up
Communication and questions	answers something else, and says "no questions" at the end	expresses clearly and asks one or two genuine questions	states reasoning when stuck, confirms the question first, and asks questions showing homework

Two threshold rules are hard, so do not relax them yourself: any dimension below 3 goes onto the weak-point list; a total below 18 means rerunning the whole process in two days' time, rather than pressing on. Two days rather than the same day leaves room for targeted practice in between; a same-day rerun only memorizes the same speech better.

Turning "did not present well" into an action for tomorrow

With the scores in, the last step translates low scores into actions. Use four fixed columns, one weak point per row: symptom, root cause, the smallest action to practice tomorrow, and how to verify. Those four columns are tomorrow's (D29's) input, so do not change the format.

The first column, symptom, is the easiest to write badly. The criterion is simple: a symptom must be a fact you can point at in the replay, with a time and the actual words. "Presented poorly" is not a symptom, and "eight minutes into the project deep dive I still had not said what I did" is; "the system design was vague" is not a symptom, and "started drawing boxes 40 seconds in and never asked a single clarifying question" is. Being unable to write such a sentence usually means you did not actually replay it and are summarizing from impression.

The second column, root cause, goes one level down to a behavioural habit or a missing preparation rather than stopping at "not practiced enough." "Eight minutes without mentioning myself" usually roots in telling the story chronologically — starting from the project's inception naturally takes a long time to reach you; "afraid to speak" usually roots in never having practiced thinking aloud and being used to thinking it through before acting. Get the root cause right and the third column almost writes itself.

The third column, smallest action, has one hard constraint: it must be one concrete thing finishable within tomorrow. "Improve my communication" is not an action, and "rewrite the three project bullets in STAR form so the first sentence out of my mouth is the conclusion" is. The fourth column, how to verify, must be an observable phenomenon, ideally with a numeric threshold, or tomorrow you still will not know whether you practiced it into place.

A completed list looks like this:

Symptom	Root cause	Smallest action to practice tomorrow	How to verify
Eight minutes into the project deep dive without saying what I did	told it chronologically from inception with no conclusion first	rewrite the three project bullets in STAR form so the first sentence is the conclusion	re-record; "the part I owned" must appear within 90 seconds, and over time is a fail
Not a word in the first 3 minutes of coding	used to thinking it through before acting, never practiced thinking aloud	take a problem I already know and narrate the reasoning throughout	replay the recording; the longest silence is under 15 seconds
Started drawing boxes 40 seconds into system design	never treated clarifying's 5 minutes as mandatory	memorize D26's five clarifying questions and rehearse the opening three times	time it; no architecture term appears before minute 5

The list needs at least 3 entries and should not be greedy. More than 5 means you are enumerating flaws rather than choosing what to practice — tomorrow is one day, and what can genuinely be practiced is usually 3. Choose by fixing the ones that are audible after one change first: the order you open with, whether there are numbers, whether there is any sound. Those three have far better returns than memorizing one more piece of knowledge.

Source Reading

Hands-On Lab

Today's output is not code, it is a rubric and a weak-point list. So there is no starter and no solution, only the recordings and two tables — and they are tomorrow's only input, so whatever you fudge cannot be practiced tomorrow.

Prepare three things before starting: a question list (self-introduction, one deep-dive question per project, one coding question, one system-design question, two behavioral questions), a countdown timer, and a recording tool. Run the domestic set in the morning and the overseas set in the afternoon, with a gap between so the second is not merely the first's speech repeated.

Run a full domestic round to the five time boxes: self-introduction 3 minutes, project deep dive 25, coding 20, scenarios and fundamentals 10, candidate questions 5. Record throughout, with no pausing, no looking things up, and no retakes.
Run the second set to the overseas five rounds: recruiter screen 30 minutes, coding 60, system design 60, behavioral 45. Write 2 to 3 reusable stories before behavioral, and apply D26's five time boxes directly to system design.
Replay the recordings, score 1 to 5 on each of the five dimensions, and total them. Beside each score write one sentence saying "because at minute N I said X"; a score you cannot justify that way does not count.
Expand every dimension below 3 into the weak-point list's four columns — symptom, root cause, smallest action, how to verify — with at least 3 entries and no more than 5.
Save the list to a file and keep it to hand. If the total is below 18, mark a rerun two days out on the calendar, and practice from the list tomorrow first.

Interview Questions

Today's three questions are in the bank below, all about process and self-assessment — the technical questions filled the previous 27 days, and today adds the "how do you prepare and review" side, which appears in recruiter screens and behavioral rounds more often than you would think. Expand a question and read the analysis before the key points.

Checklist and Tomorrow

Run through one complete China-domestic-style mock interview, from self-introduction to the candidate-questions segment
Run through one complete overseas-style mock interview, including a behavioral interview and system design
Score yourself with a rubric, and list at least 3 concrete weak points
Say what the domestic five segments and the overseas five rounds each examine, and the fundamental structural difference between them
Beside all five dimension scores, you can say "because at minute N I said X"
The weak-point list's four columns are filled in, with every smallest action finishable in a day and every verification observable
Answer at least 2 of the 3 interview questions without looking at the key points

Tomorrow (D29) works through that list item by item while warming up on four frequently asked coding problems (a rate limiter, an LRU cache, concurrency control, and streaming JSON parsing). The order is deliberate: with a list first, what you practice is the two sections you genuinely lack; without today's weak points written down, tomorrow's practice degrades into a vague review from start to finish that looks industrious and moves nothing weak.

Mock Interview Day: One Full China-Domestic-Style and One Full Overseas-Style Round, Self-Assessment