Sample verdict. Every word below belongs to someone else — yours gets your own.

Take the hot seat

Report — Payments SDE-2 · JD run

Pressed on your own numbers, you gave ground: the shape was right, the specifics never arrived.

28 July 20267 questions · HardBackend Engineer (SDE-2), Payments

Overall verdict

72/100

Technical74/100
Communication69/100
Problem solving71/100

Fix these things

ranked · biggest gain first

01

Rehearse the three résumé numbers you are proudest of until each one comes with its instrument, its window and its load in a single sentence. This is the highest-value change on the page: it is the difference between the two answers that scored worst and the two that scored best, and it costs an evening.

02

Finish every design answer by naming where it breaks

Two sentences — the failure you did not cover and how you would cover it — turns a correct answer into a senior one, and it is the exact gap between your webhook answer and the job posting.

03

Cut the pace

Read one of your own transcripts out loud at 140 words a minute and notice how much of the filler simply has nowhere to go once the sentence has room.

top 3 of 4 · the rest ranked lower

How you sounded

measured from the recording · never guessed

Pace

176wpm

Rushed

Avg pause

98ms

Fillers

5total

Pitch variation

21.6Hz

Energy

0.038rms

Mean pitch

121Hz

Steadiness

1.90% jitter

7.10% shimmer

Voice clarity

16.8dB

Uptalk

9/34rose

Articulation

4.61syll/s

Phonation

74%

Trailing off

4/18faded

Transcriber confidence

-0.31log prob

Looked at camera

71%

Smiled

13%

Head movement

8.6°/s

Slouched

15%

Hands near face

5%

Shoulder tilt

2.8°

Wrist motion

0.14sw/s

Time to first word

s

Where your pace sits

composed band · 110160
you · 176
80 · slow200 · rushed

pace & pauses ← word-level timings · pitch, energy & voice ← raw audio · on camera ← the webcam, in yourown browser · syllables & fading← intensity contour · posture← body landmarks, on device · never the transcript

Your best and worst answer

straight from your own recording

Strongest answerQ2

The consumer and the replay tool were mine. The schema change was Priya's — I reviewed it, I didn't write it.

Asked to separate yourself from the team, you did it in two sentences and gave credit away without hedging. Most candidates fold the whole project back into 'I' at this exact moment.

Weakest answerQ3

It went from about 800 milliseconds down to under 500, I think. I'd have to check the dashboard.

This is your own headline claim and it arrived softer than the résumé that made it. 'I think' and 'I'd have to check' on your strongest number invites the interviewer to discount every other number you gave.

What worked

You reach for the mechanism before the story. Asked how to make a retried webhook safe, the first thing out of your mouth was the unique constraint, not a paragraph about reliability.

the provider sends an event id, so I'd store that in a table with a unique constraint and just drop anything I've already seen

You answer the question that was actually asked. When the follow-up narrowed from 'we' to 'you', you narrowed with it instead of restating the team's work in the first person.

The consumer and the replay tool were mine. The schema change was Priya's — I reviewed it, I didn't write it.

You own a miss without decorating it. The deadline answer named the wrong assumption in one sentence and did not reach for a reorg or a dependency to blame.

The estimate assumed the schemas already matched, and they didn't.

What didn't

Your own numbers arrive without their measurement. The 40% latency win is on your résumé and you repeated it here, but you could not say what you measured it with or over what window — so an interviewer has to take it on trust, and at SDE-2 they won't.

It went from about 800 milliseconds down to under 500, I think. I'd have to check the dashboard.

fixFor every number on your résumé, be ready with three things in one breath: the tool it came from, the window it covers, and the load it was under. 'p99 over seven days in Datadog, at peak roughly 900 requests a second' — that sentence is the difference between a claim and a result.

You stop one layer above the failure. The idempotency answer got to the unique constraint and the transaction and then closed, leaving the two cases the posting actually cares about — a retry arriving while the first is still in flight, and a side effect that lives outside the database — unaddressed.

I think that covers the retries.

fixWhen you finish a design answer, say out loud where it breaks: 'this holds unless two deliveries race, which is where I'd take a row lock on the event id — and the email send is outside the transaction, so that one needs an outbox.' Naming the hole is what reads as senior; leaving it is what reads as not having looked.

Under pressure the pace climbs and the sentences run together. 176 words a minute with barely a tenth of a second between them is a candidate talking through the thinking rather than before it, and every filler you used landed in the two answers you were least sure of.

So we underestimated the migration, um, and the review cycle took longer because like coordination between the two teams needed a full sprint we never planned for

fixTake the first two seconds of every answer as silence and spend them on the shape: point, evidence, consequence. The pause costs you nothing — the interviewer is still writing — and it buys a sentence that lands instead of one that runs.

Every question, one by one

what you said · why it scored · what to say instead

You said

So we underestimated the migration, um, and the review cycle took longer because like coordination between the two teams needed a full sprint we never planned for. The estimate assumed the schemas already matched, and they didn't. I mean we did ship it, it was about three weeks late.

spokencurrent wordupcomingfiller×3 fillers
looked away3 times

Rubric

relevance

7/10

correctness

7/10

structure

6/10

depth

5/10

filler

4/10

How the answer was built

Situation 81.5%Task 0%Action 0%Result 18.5%

No TaskNo Action

Four fifths of this answer sets the scene and the outcome is one clause; nothing in between says what you were on the hook for or what you did about it.

How to improve this answer2 fixes

  1. The word 'we' is doing a lot of work in this answer. Say whose estimate it was.
  2. You named what slipped but not what you changed afterwards. A missed deadline with no changed practice reads as bad luck rather than as learning.

Possible answers

Angle 01

'We committed to six weeks on an estimate that assumed both schemas already matched. They didn't, and I found out in week three. It shipped three weeks late. What I'd change is the assumption itself — I'd have spent half a day diffing the two schemas before we gave the date.'

Angle 02

Put the correction where the miss was: 'the estimate was mine, and it was wrong because I priced the migration and not the coordination.'

Take the hot seat

free · 8 questions · no card

back to grill