Six months into a two-year re-architecture you become convinced the target architecture is wrong. Walk me through how you raise it and what you propose.

What the interviewer is testing

Not whether you can spot a bad architecture. They are testing whether you can change an organisation's mind without destroying your own credibility or the programme's. Most senior architects have been in this position, and most handle it badly in one of two directions: silence until the failure is undeniable, or an all-staff declaration that the plan is wrong, which forces everyone who backed it into defending it.

The clarifying questions that change the answer

• Wrong how? "It will not meet a stated requirement" is a different conversation from "I would have chosen differently". Only the first justifies the cost of reversing.
• What is the evidence, and is it falsifiable? A latency measurement from a slice already in production is an argument. A prediction is an opinion, and the correct move for an opinion is to go and get the measurement first.
• What has already been spent that is not recoverable? Learning, tests, extracted seams and data work usually survive a change of target. If 80% of the work transfers, this is a course correction rather than a restart, and saying so changes how it is received.
• Is there a smaller change that fixes it? Reversing the target is the most expensive option available and should be the one you argue against hardest before proposing it.

A strong answer's arc

Get the evidence before the meeting, from the slice that is already live rather than from a document. Take it privately to the person who owns the decision before it reaches a committee, framed as a risk you want to test together rather than a verdict. Bring three options with costs: continue and accept the consequence, adjust within the current target, or change the target — and state which you recommend and what would change your mind.

Then propose a decision gate rather than a decision: a two-to-four week spike with a named, falsifiable exit criterion. This is the move that separates senior from staff. It converts an argument between opinions into an experiment with a date, it lets the sponsor say yes without conceding anything, and if the spike shows you were wrong you have lost a month instead of the relationship.

Common weak answers

• "Escalate immediately." Escalation is a tool for a blocked decision, not an opening move. Used first, it teaches the people you need that you will go over their heads.
• "Sunk cost is a fallacy so ignore what has been spent." Correct in economics and useless in the room. Money spent is not a reason to continue and is very much a reason people will resist, and the way through is to show how much of it transfers.
• "Keep building and raise it at the next review." The cost of reversal grows roughly with the code written against the target. Six months is early; eighteen is often past the point where reversal is affordable at all.
• "Write a document." Necessary and not sufficient. Documents settle nothing that has a stakeholder attached.

What a strong answer adds

The organisational layer. Name who loses face and give them a way not to — frame the original decision as correct given what was known, because it usually was. Say what it costs the roadmap in weeks, not in adjectives. And state the counterfactual you will hold yourself to: if the spike does not clear its criterion, you stop asking. An architect who reverses twice without evidence does not get heard a third time.