I lost an onsite loop two weeks ago. 5 rounds at a company I genuinely wanted to work at, all the way through to the verbal "we want to send an offer," and then 6 days of silence followed by a one-line rejection. Standard pattern from the 142 rejection emails I read recently.
I asked the recruiter for a real debrief. To his credit, he gave me one. Twenty minutes on the phone, no marketing speak, real specifics about why I lost. I have been thinking about it for two weeks because there are 5 lessons in it that I had not internalized before, and they are probably useful to share.
What happened, briefly
5-round loop: recruiter screen, hiring manager call, technical pair, system design, behavioral with two ICs and a director. After round 5 the recruiter said "we want to move to offer discussion, expect to hear by end of week." Wednesday. By the following Tuesday: silence. The Tuesday after that: rejection.
The recruiter was honest in the debrief: there was a stronger candidate in the loop they had not closed yet, and they wanted to wait to see if that candidate accepted. When that candidate did, mine became a polite no.
That part is not the lesson. That part is the new normal, and you cannot do anything about it.
The lesson is what was technically wrong with my candidacy that made me the backup, not the primary. Five things.
Lesson 1: The system design round needs an "I have actually done this" anchor
The recruiter said my system design round was "good but textbook." I designed a notification service the way the textbook teaches. Queues, fanout, dead letter handling, retry logic, the whole bit. The interviewer's note was: "candidate clearly knows the patterns but did not surface a single time he had built or maintained one of these in production."
The fix, in retrospect: at the start of the design, anchor the answer in something I actually shipped. "I have built two of these — one for billing notifications at Helio, one for incident pages at the previous job. The Helio one looked like this; let me start there and work toward your requirements." That immediately reframes the round from "is the candidate familiar with the textbook" to "is the candidate experienced enough to extend a real system."
I knew this. I just did not do it.
Lesson 2: The behavioral round wants a story arc, not a bullet list
Two of the behavioral interviewers told the recruiter that my STAR-format answers felt "rehearsed and shallow." I was using STAR exactly the way every interview prep book says to. Situation. Task. Action. Result. Two-minute answer.
Their note: the candidates who got the highest ratings told stories — with stakes, with surprise, with specific people, with the moment the plan failed and what they did about it. The story had a Situation, a Conflict, and a Resolution, but the framing was narrative, not bulleted.
The fix, in retrospect: stop trying to "complete" the STAR. Tell the actual story. "We were two days from launch and our load test failed at 18% of expected traffic. The CTO wanted to ship anyway. I disagreed publicly in the room, and here is what happened next." That is a story. The bullet-list version was technically the same content but read as canned.
Lesson 3: The pair-coding round wants me to think out loud, not just code
The technical pair went well from my perspective. Solved the problem in 35 minutes, both test cases passed, used the right data structure. The interviewer's note: "candidate was almost completely silent during problem-solving phase. Hard to evaluate thinking process."
This one stung because I knew it. Quiet solving feels efficient to me; it reads as inscrutable to the interviewer. The interviewer cannot give credit for thinking they cannot hear.
The fix: narrate one sentence per minute, minimum. "I am going to start with the brute force, then look for a hash-based optimization." "I think the edge case here is empty input; let me handle that first." Even unnecessary narration beats silence.
Lesson 4: The hiring manager call wants a specific question that proves I read about the team
I asked smart questions. I asked about the team's roadmap, their on-call rotation, what the manager looked for in their best ICs. The manager's note: "candidate's questions were generic and could have been asked of any team."
The fix: ask one question that demonstrably required research. "I noticed your team published a blog post 6 months ago about migrating from X to Y. How did that play out — is the migration complete, and how does that change the priorities for the next 6 months?" That kind of question signals the candidate cares about this team specifically. Generic questions signal the candidate is interviewing in volume.
Lesson 5: The recruiter screen sets the comp anchor — be ready
This was the most actionable lesson. The recruiter asked early, almost casually, "what comp range are you targeting for this role?" I gave a range. The range was $20K below the band's median.
Their note: "anchored low, did not push back when I asked." I had handed them the lower half of the band as my expectation, before any leverage existed in the conversation.
The fix: do not give a number first. "I would like to keep comp in scope until I better understand the role and the leveling on your side. I am sure we can find a range that works. Can you tell me about your leveling matrix and where this role fits?" That is the textbook deflection and it works. Then, if forced to give a number, give a target near the top of the public band for the role.
What I am taking forward
The five fixes above are concrete enough that I can practice them. None of them require me to become a better engineer. They require me to perform better as a candidate.
I am also no longer treating the rejection as personal. The recruiter told me directly: I was second. The other candidate was first. Headcount was one. That is the most common pattern in the 5 patterns I wrote about last week, and it does not mean my candidacy was bad. It means there was one slot.
For the prep side of an upcoming onsite, the day-by-day plan from 4 recent loops is what I am going to use to drill the 5 fixes above into the next loop.
The bigger lesson: ask for the debrief. They mostly say no. But when they say yes, you get 20 minutes of feedback that no amount of self-reflection would have produced.
Top comments (0)