Feedback fails for a predictable reason: it usually arrives as a verdict about the past instead of a question about the future. 'This section is wrong' closes a person down because there's nowhere for them to go except defend it or concede it. Neither gets you a better result.
The fix is mechanical, not emotional. Replace the verdict with a how question aimed forward. Instead of 'this analysis is wrong,' try 'how would this hold up if a customer challenged the assumption on line four.' You're pointing at the same flaw, but you've handed them the pen instead of taking it from them.
This works because 'how' questions require thinking, and thinking and defending don't run at the same time. A person answering a how question is doing math in their head. A person answering a why question is building a case for themselves. You want them doing math.
Save the direct correction for things that are genuinely non-negotiable, like a compliance error or a factual mistake with real cost attached. For everything else, a good how question gets the person to find the flaw themselves, which they will defend far less than a flaw you found for them. People rarely argue with their own conclusions.
A verdict closes someone down. A forward-looking "how" question gets them thinking instead of defending, and they usually find the flaw themselves.