After a full review the CTO chooses an approach you argued against. You believe it will cause serious problems in eighteen months. What do you do?

Separate the two questions

Was the decision made properly? Were the options genuinely considered, was your argument understood, was the trade-off stated? If yes, the process worked and you lost, which is a normal outcome and not a failure.

Is it a risk you can live with, or a line you cannot cross? Almost all disagreements are the first. A small number — a genuine safety, legal or security exposure — are the second, and it is worth knowing in advance which one you are in, because the response differs completely.

For the ordinary case, which is nearly all of them

Record the decision and your dissent, without drama. An ADR stating the decision, the options, the reasoning and — in the consequences section — the risk you raised and the fact that it was accepted. Not a note saying you disagreed; a technical statement of what may happen and under what conditions.

That serves three purposes: the next person understands the reasoning, the risk is in a register rather than in your memory, and if it materialises the conversation is about a known accepted risk rather than about who was right.

Then commit properly. Half-hearted implementation of a decision you opposed guarantees the outcome you predicted and makes it impossible to tell whether the decision or the execution failed. Build it as well as it can be built.

Define the trigger. Agree in advance what evidence would indicate the risk is materialising — a specific metric, a specific threshold — and when it will be reviewed. That converts "I think this is wrong" into a testable claim with a date, which is far more useful to everyone including you.

For the rare case

If it is genuinely a line — a real safety, legal or security exposure — say so explicitly, in writing, once, to the person accountable. Not as a threat and not repeatedly. Then it is their decision with full information, which is the most an architect can do and is where your responsibility ends.

What not to do

Do not relitigate it in every subsequent meeting. Do not build a parallel design. Do not say "I told you so" if it goes wrong — the record already says it, and saying it out loud spends the credibility that would otherwise let you win the next argument.