Kwame

Cost and latency belong in the same spreadsheet as quality, because a switch that improves output slightly and doubles response time is not an improvement in a product people use interactively. I have made that trade without noticing I made it.

It feels better is not nothing, it is just not evidence, and it is indistinguishable from novelty.

Switched on release day twice and regretted both. What the fixed set of real inputs gives you that a feel does not: it catches the regression on the case you do not think to try, which is always the awkward input a customer sends once a month.

Twenty real cases and an afternoon is genuinely enough to stop this. The blind comparison is what stops you scoring the one you already prefer.