Lakshya

Handbook · Chapter 5

Discovery before solution

The decisive round in three separate archetypes, with the lowest pass rate of any stage.

The Forward Deployed case study, the Solutions Architect customer presentation and the Platform Lead strategy round are the same exam. A vague problem is described, and you are scored on what you do in the first ten minutes. The FDE version is reported at roughly a 40% pass rate and around 30% of total loop weight — the lowest and heaviest single stage anywhere in this atlas.

the prompt “model is wrong 30%” minute 3 propose a fix fine-tune / better prompts solves the stated problem which was not the problem ask what “wrong” means who measured it, on what “no one measured it” the real deliverable is an eval Both candidates are technically competent. Only one of them found out what the job was.
Both candidates are competent; they diverge at minute three. One proposes a fix to the problem as stated. The other establishes what “wrong” means, who measured it and on what data — and discovers there is no measurement, which makes the real first deliverable an evaluation rather than a model change.

Why competent people fail this round

Because everything else in an engineering career rewards the opposite instinct. You are hired to solve things; solving quickly is usually good; and a candidate sitting in silence feels like a candidate who does not know the answer. So the strong engineer produces a plausible solution at minute three and spends the remaining fifty minutes elaborating it — competently, and in the wrong direction.

The reframe that makes this easy

You are not being tested on whether you can solve the problem. You are being tested on whether you can find out what the problem is. Those look similar from the inside and score completely differently.

The practical tell: if you have proposed anything before you know how the customer measures success, you are already in the failing path.

A discovery bank

Learn six, not fifteen

Have a set you can reach for under pressure. These transfer across all three archetypes:

  • “How are you measuring that today?” — the highest-yield question in the set, because the answer is very often “we aren't”, and that changes the entire engagement.
  • “Who noticed this, and what did they see?” — separates the reported problem from the experienced one.
  • “What have you already tried?” — stops you proposing something that failed last quarter.
  • “What happens if we do nothing?” — establishes real urgency and real value, and occasionally reveals there is none.
  • “Who has to agree before this ships?” — surfaces the security review, the procurement gate, the VP who was not in the room.
  • “What does success look like in ninety days?” — converts a vague ask into an acceptance criterion you can hold them to.

Then scope, out loud

Discovery without a plan is just interviewing the interviewer. Once you understand the problem, narrate the decomposition: what you would do in week one, what you would deliberately not do, what would make you stop. Saying “I would cut the multi-language support for the first release” scores; quietly not mentioning it does not.

The drill

Unpleasant and highly effective

  • Take a one-line business problem. Set a timer for forty-five minutes. Talk through it to a recording — discovery, reframe, scope, risks, what you would cut.
  • Watch it back. Mark the timestamp where you first proposed a solution. If it is before minute eight, that is your entire finding.
  • Repeat three times with different problems. Most people move that timestamp from minute two to minute ten within three attempts, and that alone changes the outcome of the round.
  • “A customer says the model is wrong 30% of the time. What do you do first?”Ask what “wrong” means and how it was measured. Do not propose anything yet.
  • “You have two weeks and an executive demo. What do you cut?”A scoping test wearing a delivery costume.
  • “Their data is far worse than they told us. Walk me through that conversation.”Deliver bad news without sugar-coating and without blaming. Bring the replan with you.
← Latency and cost arithmeticFailure and incident narrative →