Technical disagreement interview question: when to revisit the decision

“What would make you revisit the decision?” is a follow-up to a behavioral interview question about a technical disagreement. A concrete answer identifies the assumption behind the team’s choice and the evidence that would put it in doubt.

Amazon’s senior software engineering interview guide asks candidates to explain the reasoning behind past decisions. Its leadership principles describe challenging decisions respectfully and supporting the chosen approach once the team decides. Amazon SDE III interview preparation, Amazon Leadership Principles

Why the team chose a direct download

In this fictional example, an engineering team is adding CSV exports to an admin dashboard. The candidate proposes a background job. The browser would request an export and check its status until the file was ready. The team chooses to generate the CSV during the download request. That version can ship sooner, and test exports at the expected pilot size finish within the gateway’s 30-second timeout. A background job would add a worker and a way to track export status. Azure’s asynchronous request-reply pattern

“I added a test at the pilot’s maximum export size and logged the time spent waiting on the database separately from the time spent building the CSV. Before launch, we agreed to pause expansion if at least one percent of export requests timed out. The direct download shipped to the pilot on schedule.”

Same timeout rate, different evidence

The interviewer adds a detail. “A week later, 30 of 1,000 export requests timed out. Would you reopen the decision?”

The 3% timeout rate crosses the 1% threshold for pausing the rollout. The timing logs can help distinguish two reasons for the same number.

The timeouts cluster around a database failover

The timing logs show that the failed exports spent most of the request waiting to reconnect to the database. Requests at the same export size finish well within the deadline once the database is healthy.

“I’d keep the rollout paused while we check recovery from the failover. The direct download still fits the deadline at the size we agreed to support when the database is healthy. I’d fix or verify the recovery behavior, then rerun the pilot checks before expanding.”

The largest supported exports repeatedly hit the deadline

The database is healthy. At the top of the supported range, timing logs show the server still building the CSV when the gateway closes the request. The team can reproduce it with representative data and load.

“I’d take that back to the team. We chose direct downloads because exports within the pilot limit were expected to fit the deadline, and we can now reproduce cases where they don’t. I’d check whether we can bring generation time within the deadline or agree on a smaller export limit that still meets users’ needs. If users need files that take longer to build, I’d compare that with moving generation to a background job.”

A background job lets CSV generation continue after the initial request ends. Users would still wait for the file. That tradeoff is worth comparing if the export sizes users need routinely exceed the request deadline.

Practice the follow-up

Choose a real disagreement with two plausible approaches. Describe why the team picked one, what work went into shipping it, and which observation would warrant revisiting it. If the team had no threshold, name the signal that would have been useful at the time.

For a practice round, give a partner one symptom with two possible causes. Explain what to check, what to do for affected users, and whether each result changes the decision. How to explain your project in an interview