Differentiators, and interview-day mechanics

What it is

Two things that sit either side of the interview itself: the small number of behaviours that make a director remember you, and the physical and logistical mechanics that decide whether you perform at your actual level on the day.

DIFFERENTIATORS, roughly by how few candidates do them:

  bring an artifact                       almost nobody
  send a 30/60/90 plan after the onsite   almost nobody
  know their public engineering output    few
  speak in operational metrics            few
  name your own gap before they find it   few
  hold a defended technical opinion       some
  send a follow-up that ADDS something    some
  prepare the questions you would ask the
    team in week one                      almost nobody

MECHANICS, which decide whether the above lands:

  whiteboard and remote setup
  the order you draw things in
  stamina across a five-hour loop
  what you do in the ten minutes between rounds

What this is confused with: differentiators as flourishes. They are cheap and they are decisive at the margin, which is where most staff decisions live: a loop where four interviewers say "hire" and one says "not sure" is decided by whatever the hiring manager remembers.

Also confused: mechanics as trivia. A candidate who is exhausted by round four performs a level below themselves in the two rounds that are usually the leadership ones, and those are the rounds a staff decision turns on.

The problem it solves

Most staff loops end with a set of positive assessments that are hard to distinguish.

Five candidates, all technically adequate, all with good
stories. The debrief:

  "Strong. Good design round."
  "Solid. Nothing concerning."
  "I'd hire."
  "Good communicator, deep on the storage question."
  "Strong. I liked them."

The hiring manager has to pick. What they pick on is whatever
is MEMORABLE and SPECIFIC, and in the absence of anything
memorable they pick on rapport, which is the mechanism by
which loops select for similarity.

An artifact, a 30/60/90 plan, a defended opinion and a
follow-up that continues the design discussion are four
memorable, specific things, and they cost a total of about
four hours.

And the mechanics problem is measurable in a way people do not expect:

A five-round onsite, 09:00 to 15:00.

Rounds 1 and 2: sharp.
Round 3: fine.
Rounds 4 and 5: noticeably slower, more hedging, worse
  recall of numbers.

Rounds 4 and 5 at most companies are the leadership and
hiring-manager rounds, which is where a staff decision is
made.

The candidate's ability did not change. Their blood sugar,
hydration and cognitive load did, and none of that was
planned for.

Mechanics

The eight differentiators

1. Bring an artifact.

A one-page architecture diagram of something you built,
sanitised. Or a design doc you can share. Or the
before-and-after of a migration.

Why it works: the deep-dive round is you describing a system
from memory while an interviewer tries to hold it in their
head. A diagram removes that load entirely and changes the
conversation from reconstruction to discussion.

Practical: bring it as a PDF you can screen-share and as a
paper copy if in person. Sanitise properly, remove customer
names, internal hostnames and anything under NDA, and SAY
that you have sanitised it, which is itself a signal about
judgment.

Almost nobody does this.

2. Send a 30/60/90 plan after the onsite.

One page, for THEIR team, sent within 48 hours.

  FIRST 30: listen and map. Name what you would read, who you
    would meet, what you would measure. Specificity is the
    whole thing: "read the last two quarters of incident
    reports and the on-call handover notes" beats "get up to
    speed".
  60: one visible fix. Something small, real and derived from
    what you learned in the loop.
  90: a proposal. A direction, with the evidence you would
    have gathered by then to support it.

Include the caveat that makes it credible: "this is written
from the outside, so half of it is probably wrong; the point
is how I would approach the first quarter."

For a lead role this is close to decisive, and the reason is
that it is the only artifact in the whole process that shows
what you would DO rather than what you have done.

3. Know their public engineering output.

Their engineering blog, conference talks, open source, RFCs.

Reference it SPECIFICALLY: "in your post on the sharding
migration you mentioned the dual-write window ran longer than
planned; how did that land?"

Two hours of reading. It changes the interviewer's model of
you from "candidate" to "someone who is already interested in
our problems", and the specific reference is what does it: a
general "I read your blog" does the opposite.

4. Speak in operational metrics.

p99, error budget, DORA metrics, cost per query, on-call
pages per week, toil percentage, cycle time.

A lead who talks about cost and toil sounds senior
immediately, because those are the two things engineers
under-report and managers care about.

  weak:   "we improved performance"
  strong: "p99 went from 800ms to 120ms, which mattered
           because checkout conversion moves measurably below
           about 400ms, and it cut our compute bill by 22%
           because we stopped over-provisioning for the tail"

5. Name your own gap before they find it.

"The gap in my background is that I have never run a team
 through a reorganisation, and here is how I would approach
 it in the first quarter."

Why it works: the gap is going to be found. Naming it first
converts a concern into evidence of self-awareness, and it
lets you frame the mitigation rather than defend against the
objection.

The failure mode is naming a fake weakness ("I care too
much"), which is worse than saying nothing.

6. Hold a defended technical opinion.

Not a hot take. A position, in their domain, with the
reasoning and the conditions under which you would change it.

  "I think most teams adopt a service mesh a year too early.
   The cost is per-pod memory and a debugging surface nobody
   has skills for, and the benefit only exceeds a library
   approach once you have more than about two languages or a
   policy that changes faster than code ships. I would ask
   which of those is true here."

Directors hire people with a point of view, and the
reversal condition is what distinguishes an opinion from an
attitude.

7. Send a follow-up that adds value, not thanks.

Within 24 hours, because the debrief is fast.

  weak:   "Thank you for your time, I enjoyed our
           conversation."
  strong: one paragraph continuing the design discussion, or
          the diagram you sketched on the whiteboard, cleaned
          up and attached.

And the highest-value version: if you asked whether anything
about your background gave them hesitation, this is where you
answer it. If you flubbed something and know it, correct it
in one line: "I said X for the quorum math; on reflection
it's Y." That correction has rescued candidacies, because it
demonstrates exactly the self-correction the loop is trying
to measure.

8. Prepare the questions you would ask the team in week one.

A short list, shared at the end of the hiring-manager
conversation or in the follow-up:

  "What is the thing everyone knows is broken and nobody has
   time to fix?"
  "When were you last paged at night, and what was it?"
  "What decision is currently stuck?"
  "What would you want a new lead to NOT change in the first
   month?"

Sharing this signals that you already think like the lead,
and it is the cheapest of the eight.

Whiteboard and remote mechanics

DRAWING ORDER, which matters more than drawing quality:

  1. the USER and the request, at the left
  2. the boxes, left to right, in request order
  3. the data stores, below the services that own them
  4. the arrows, labelled with the PROTOCOL and the
     approximate volume
  5. only then the detail

Reserve space before you start. The most common whiteboard
failure is running out of room and redrawing, which costs
three minutes and the interviewer's attention.

WRITE THE ASSUMPTIONS IN A CORNER and keep them visible:
QPS, data size, latency target, consistency requirement.
You will refer back to them, and an interviewer who can see
them can follow your reasoning.

SAY WHAT YOU ARE DRAWING while you draw it. Silence with a
marker is the least informative state in the interview.

REMOTE SETUP, and this is a preparation task, not a
last-minute one:

  - test the tool THEY use, not the one you like. Excalidraw,
    Miro, CoderPad and a shared Google Doc all behave
    differently and the first two minutes of a round is a
    bad time to learn one.
  - a second monitor, so the diagram and the faces are both
    visible
  - a wired connection if at all possible
  - camera at eye level, light in front of you rather than
    behind
  - your notes on paper, not on screen, because reading from
    a screen is visible and reading from paper is not
  - a glass of water and something to eat, in reach
  - phone silenced, notifications off at the OS level,
    including calendar popups that will show your other
    interviews

PRACTISE DRAWING IN THE TOOL. Diagramming fluency in
Excalidraw is a skill with a learning curve, and spending it
during the round costs you the design round.

Stamina

A five-round onsite is four to six hours of sustained
performance, which almost nobody trains for.

  EAT BEFORE, and eat something in the middle. A 45-minute
    round at hour five on no food is measurably worse.
  WATER, continuously.
  THE TEN MINUTES BETWEEN ROUNDS: stand up, look out of a
    window, do not review notes. Reviewing notes between
    rounds raises anxiety and changes nothing, and the
    recovery is worth more.
  RESET DELIBERATELY. Each round has a new interviewer who
    has not seen the previous one. A bad round is not
    carried forward unless you carry it. Say to yourself,
    explicitly, that the last round is over.
  THE FIRST MINUTE of each round: ask what they would like to
    cover and how long they have. It buys you thirty seconds
    of orientation and it makes you look organised.

SCHEDULE, where you have a choice: not first thing Monday,
not last thing Friday, not the day after a release. And put
NOTHING else in that day.

IF A ROUND GOES BADLY, say so once at the end, briefly, and
move on: "I don't think I did that problem justice; I'd
approach it differently with more time." Then reset. Nothing
is gained by carrying it and something is lost.

A worked example: two candidates, one memorable

Two candidates at final stage for a team lead role, both assessed as technically capable, both with good stories. The hiring manager had one headcount.

Candidate A's loop:

Five rounds, all "hire" or "strong hire". Good design round,
solid deep dive, competent leadership answers.

Follow-up email, same day: three sentences of thanks.

Debrief comments:
  "Strong. Good communicator."
  "Would work well with the team."
  "Hire."

Candidate B's loop:

Five rounds, four "hire" and one "strong hire". Marginally
weaker on the coding round.

BUT:

Brought a sanitised one-page architecture diagram to the deep
dive, screen-shared it, and said "I've removed the customer
names and internal hostnames; happy to go into anything on
here."
  -> the deep dive spent 30 minutes on the diagram's
     failure-handling section rather than 15 minutes on the
     candidate reconstructing the system from memory

Referenced the company's engineering blog specifically:
"your post on the ingestion rewrite mentioned the backfill
took three weeks longer than planned; was that the schema
inference or the throughput?"
  -> the interviewer, who had written the post, spent ten
     minutes on it and mentioned it in the debrief

Named their own gap unprompted: "I've never run a team
through a reorg, and if that's a concern here I'd want to
talk about how I'd approach it."
  -> the hiring manager's note: "raised the thing I was
     going to ask about, before I asked"

Held a defended opinion on a technology the team was
evaluating, including the condition that would change their
mind.

Sent, 36 hours after the onsite:
  - a one-page 30/60/90 plan for the team, with the caveat
    that half of it was probably wrong
  - the whiteboard diagram from the design round, redrawn
    cleanly
  - one paragraph continuing a disagreement from the design
    round, conceding a point the interviewer had made

Debrief comments:
  "Brought a diagram. Nobody does that."
  "The 30/90 plan is the first one I've received. Half of
   it is wrong and the half that's right is the half I'd
   have wanted them to notice."
  "Raised their own gap before I did."

Candidate B was offered, and the hiring manager's stated reason was the 30/60/90 plan.

The cost accounting, which is the point:

diagram (sanitising an existing one)        ~1.5 hours
reading their engineering blog              ~2 hours
30/60/90 plan                               ~2 hours
follow-up email with the redrawn diagram    ~45 minutes
-------------------------------------------------------
                                            ~6 hours

Against roughly 60 hours of technical preparation, and it
was the deciding factor between two candidates the loop
could not separate.

And the honest part, recorded by the hiring manager afterwards:

"A was probably the slightly better engineer. B showed me
what they would do with the job, and A showed me what they
had done with the last one. For a lead role that is the
question I was actually asking, and only one of them
answered it."

The stamina finding, from the same candidate's earlier failed loop:

Four months earlier, B had failed a loop at a different
company. Their own reconstruction:

  09:00 coding        strong
  10:00 coding        strong
  11:00 system design good
  12:00 (no break; lunch was a 30-minute call with the
        recruiter)
  13:00 deep dive     "I couldn't remember numbers I know
                       cold. I said 'roughly' four times."
  14:00 leadership    "I rambled. I could hear myself
                       rambling."

Debrief: strong on the technical rounds, weak on the two
that decide a staff hire.

What changed for the next loop: ate before, ate at the
break, refused the working lunch, stood up between every
round, and did not look at notes after 09:00.

The two afternoon rounds in the successful loop were the two
strongest scores.

"The two rounds that decide a staff hire are the two you are most tired for" is a scheduling fact, not a performance one, and it is fixable with food, water and a refusal to take a working lunch.

Production evidence

Amazon's Bar Raiser process and Google's hiring committee both make the written debrief the decisive artifact, which is the structural reason a memorable, specific detail outweighs a marginally stronger performance: the committee reads what was written down, and "brought a diagram" is written down.

Structured-interview research consistently finds that in the absence of distinguishing evidence, evaluators fall back on similarity and rapport, which is the mechanism a differentiator interrupts: it supplies a specific, non-rapport reason to prefer one candidate.

The 30/60/90 plan is standard practice in sales and executive hiring and is unusual in engineering, which is precisely why it lands: it is the only artifact in the process that demonstrates what the candidate would do rather than what they have done.

Cognitive fatigue over sustained performance is well established: sustained demanding cognitive work degrades working memory and self-monitoring over hours, and the degradation is reduced by breaks, hydration and food. The practical consequence is scheduling rather than psychology: the rounds late in a loop are typically the behavioural and hiring-manager rounds.

Follow-up correction rescuing candidacies follows from the debrief timing: independent written feedback is typically submitted within a day, so a correction that arrives inside that window can change the written assessment, and one that arrives after the decision cannot.

Sanitisation as a judgment signal is real in regulated and enterprise contexts, where a candidate sharing an unsanitised internal document is a demonstrated confidentiality risk, and saying explicitly what you removed is the behaviour that distinguishes the two.

The debate

Are differentiators gimmicks? They would be if they were empty, and each of the eight supplies real information: an artifact removes reconstruction load, a 30/60/90 shows how you would approach the job, a defended opinion with a reversal condition shows how you reason. The gimmick version is the same behaviour without content, a generic plan or a diagram of nothing, and interviewers detect it immediately.

Is a 30/60/90 plan presumptuous? With the caveat, no. "This is written from the outside, so half of it is probably wrong; the point is how I would approach the first quarter" removes the presumption and keeps the signal, and in the worked example the hiring manager's comment was that the half that was right was the half they would have wanted the candidate to notice.

Should you name your own weakness? Yes, if it is real and if you have a mitigation. Naming a fake weakness is worse than saying nothing, and naming a real one before the interviewer finds it converts an objection into evidence of self-awareness. The judgement call is which weakness, and the answer is the one they are going to find anyway.

Is bringing an artifact risky for confidentiality? It is a demonstration of judgment in either direction. A sanitised diagram with an explicit statement of what was removed is a positive signal; an unsanitised internal document is a disqualifying one, and the difference is entirely in the preparation.

Does interview-day logistics really matter? Measurably, and the specific mechanism is that the rounds late in a loop are the behavioural and hiring-manager rounds, which is where a staff decision is made. A candidate who is sharp for the two coding rounds and vague for the deep dive has optimised the wrong half of the day, and it is fixed by food, water, breaks and refusing a working lunch.

Should you review notes between rounds? No. It raises anxiety, changes nothing, and consumes the recovery time that has a measurable effect. Stand up, look at something far away, and let the previous round end.

Follow-up Q&A

"What actually makes a director remember a candidate?"

Something specific and non-rapport, because in the absence of distinguishing evidence evaluators fall back on similarity. The eight that work, roughly in order of how few candidates do them: bring a sanitised artifact; send a one-page 30/60/90 plan for their team after the onsite; reference their published engineering work specifically; speak in operational metrics like p99, error budget, cost per query and on-call load; name your own gap before they find it; hold a defended technical opinion with a reversal condition; send a follow-up that continues the discussion rather than thanking them; and prepare the questions you would ask the team in week one.

"Why is a 30/60/90 plan close to decisive for a lead role?"

Because it is the only artifact in the entire process that shows what you would do rather than what you have done, and for a lead role that is the question actually being asked. One page: first 30 days listen and map, naming what you would read and who you would meet and what you would measure; 60, one visible fix derived from what you learned in the loop; 90, a proposal with the evidence you would have by then. Include the caveat that it is written from the outside and half of it is probably wrong, which removes the presumption. In one case the hiring manager's stated reason for the offer was the plan.

"What makes a follow-up email worth sending?"

Adding something. A generic thank-you is read and ignored. A paragraph continuing the design discussion, the whiteboard diagram redrawn cleanly, or an answer to a concern the interviewer raised is evidence your advocate can use, and it must arrive within about 24 hours because the debrief is fast. The highest-value version is a one-line correction if you know you got something wrong: "I said X for the quorum math; on reflection it is Y." That demonstrates exactly the self-correction the loop exists to measure.

"How should you use a whiteboard?"

Reserve space before you start, because running out of room and redrawing costs three minutes and the interviewer's attention. Draw in request order: the user at the left, then the services left to right, data stores below the services that own them, then arrows labelled with protocol and approximate volume, and only then detail. Write the assumptions in a corner and keep them visible, QPS, data size, latency target, consistency requirement, because you will refer back to them and an interviewer who can see them can follow your reasoning. And narrate while drawing: silence with a marker is the least informative state in the interview.

"What is the most under-prepared part of a remote loop?"

The tool. Test the one they use rather than the one you like, because Excalidraw, Miro, CoderPad and a shared document all behave differently and the first two minutes of a design round is a bad time to learn one. Diagramming fluency has a learning curve and spending it during the round costs you the round. Then the physical setup: a second monitor so the diagram and the faces are both visible, notes on paper rather than on screen because reading from a screen is visible, and notifications disabled at the operating-system level including the calendar popups that would show your other interviews.

"Why does stamina matter more than people think?"

Because the rounds late in a loop are usually the behavioural and hiring-manager rounds, which is where a staff decision is made, so a candidate who is sharp for the morning coding rounds and vague by the afternoon deep dive has optimised the wrong half of the day. One candidate's reconstruction of a failed loop: strong on two coding rounds, then a 30-minute recruiter call instead of lunch, then "I couldn't remember numbers I know cold" in the deep dive and audible rambling in the leadership round. The fix was food, water, standing up between rounds, refusing the working lunch, and not reviewing notes, and in the next loop the two afternoon rounds were the strongest scores.

Common misconceptions

"Differentiators are gimmicks." Each supplies real information. The gimmick version is the same behaviour with no content, which interviewers detect immediately.

"A 30/60/90 plan is presumptuous." With the caveat that it is written from outside and half is probably wrong, it is the only artifact showing what you would do rather than what you have done.

"A thank-you email is the follow-up." It is read and ignored. A paragraph continuing the discussion, a cleaned-up diagram, or a correction is evidence that changes a debrief.

"Whiteboard quality matters." Order and legibility matter. Running out of space and redrawing is the common failure, and narration matters more than neatness.

"I'll set up the remote tooling on the day." Diagramming fluency has a learning curve, and paying it during the design round costs you the design round.

"Reviewing notes between rounds helps." It raises anxiety, changes nothing, and consumes the recovery that measurably affects the afternoon rounds.

Interview delivery note

Say this verbatim when opening a deep dive: "I've brought a one-page diagram of the system, sanitised, so I've removed customer names and internal hostnames. Happy to go into anything on it." It removes the interviewer's reconstruction load, changes the round from recall to discussion, and the sanitisation sentence is itself a judgment signal. Almost nobody does it.

The senior-versus-staff separator is the 30/60/90 plan sent after the onsite. A senior candidate describes what they have done. A staff or lead candidate sends one page describing what they would read, who they would meet and what they would measure in the first thirty days, one visible fix by sixty, and a proposal by ninety, with an explicit caveat that half of it is wrong from the outside. It is the only artifact in the process that answers the question a hiring manager is actually asking, and in one case it was the stated reason for the offer between two candidates the loop could not separate.

The second signal is treating the day as a performance to be scheduled. Knowing that the rounds late in a loop are the ones that decide a staff hire, and therefore eating, hydrating, standing up between rounds and refusing a working lunch, is not a soft concern: one candidate's failed loop was strong on the two morning coding rounds and weak on the deep dive and leadership rounds, and the difference in the successful loop was food and breaks.

The Staff & Team Lead Interview Reference