HardBehavioralSystem design

You are the founding engineer at the end of your first week on a fictional report product. The agreed pilot outcome is that valid accepted exports produce correct, account-authorized results within five minutes. The following complete observation window covers four pilot accounts and 12 distinct accepted exports: eight completed correctly within five minutes, two completed correctly after five minutes, and two failed. Both failed jobs used the documented, supported format B; you reproduced a parser mismatch with a synthetic B fixture. No other failure cause has been established.

1Give yourself 5 minutes

2Answer out loud, not in your head

3Then compare with the answer below

The problem

You traced the request/job records and built the reproducing fixture. A teammate implemented and deployed a separate retry-cap change. A colleague checked the ten completed outputs and their account access. The provider also increased its quota that week. An earlier local run used a smaller dataset, so no comparable before/after speed result exists.

Three engineer-days are available for the next milestone. Estimates are: two days to repair format B and add focused regression checks, including review; half a day to run an authorized, bounded pilot verification after that repair; half a day to review outcomes and write the next decision; or eight days to replace the queue architecture. The format repair does not change stored-data representation; the previous parser path can remain available for rollback. Real workload volume and longer-term reliability are still unmeasured.

Write a short ownership update and choose the next milestone. Separate your contribution from others', calculate the observed timely-completion measure, identify what is known versus unverified, and specify success and stop conditions for a plan that fits three days.

Reference answer

Attribute tracing and reproduction to yourself, retry-cap implementation/deployment to the teammate, and output/access verification to the colleague. Do not claim sole delivery or credit the retry cap for all observed outcomes: provider quota also changed, and there is no comparable baseline. Report eight timely correct outcomes out of 12 accepted exports, about 66.7 percent; ten of 12 eventually completed, about 83.3 percent, which is a different measure. Both accepted format-B failures remain in the timely denominator.

Choose the two-day format-B repair, half-day bounded pilot verification, and half-day review, totaling three days. This targets a reproduced defect in a documented supported input; the eight-day queue replacement neither fits capacity nor follows from established queue evidence. Require the B fixture and supported-format regression cases to produce correct results, then verify a small authorized pilot sample through actual result retrieval and account scope. Record timings and failure outcomes under one defined window. Stop rollout or further exposure if the repair changes correct outputs, crosses account boundaries, or creates new failures; retain the previous parser path under the supplied compatibility assumption and investigate before expanding.

State remaining limits: the controlled fixture establishes the parser defect and repair behavior, but the bounded pilot does not prove a long-run service level, representative peak-load capacity, or the causal impact of the separate retry-cap change. Use the final half-day review to decide the next investigation from actual failure, latency, and use evidence, rather than promising a rewrite in advance.

Then expect these follow-ups

If the B fixture passes but a pilot output differs, what should stop and which evidence should be preserved?
If the repair estimate expands beyond two days, how would you revise the commitment without hiding verification time?
Which additional observation would justify investigating queue architecture?
Optionally, compare this decision with a real experience, clearly separating that experience from the fictional case.

Free to read · better with Enzo

Practice this out loud with Enzo

Enzo runs it as a mock interview, pushes back with follow-ups, and grades you on the rubric.

Continue with Enzo Sign up free