A strong answer. “First I'd interrogate my own study honestly, because contradicting a committed decision has a high evidential bar and it should. How many participants, how were they recruited, could the finding be an artefact of my task design? If I found this in five sessions with people recruited from our own power users, I should say so.

Then I'd separate what I know from what I infer. 'Eleven of fourteen participants could not complete the primary task without help' is an observation. 'We should not build this' is a recommendation, and mixing them is how research gets dismissed as opinion.

Then I'd go to the decision-maker privately, before any group setting. Nobody changes their mind in a room with their peers watching. And I'd frame it as risk rather than as error: 'The plan may still be right, and here's what would have to be true for it to work. My data suggests it isn't. Here's the cheapest way to find out before we spend two quarters.'

Then I'd offer a path that isn't 'stop'. Ship it to 5%, instrument the specific behaviour, and set the threshold now. Leadership will accept a test far more readily than a reversal, and a test they agreed to in advance is one they'll act on.

And if they proceed anyway — which happens and is sometimes correct, because they know things I don't — I'd write it down with the prediction and the date. Not to be proved right, but because the second time research is ignored and turns out to have been right, the third study gets taken seriously.”