The evidence that was inadmissible

A confession obtained without a lawyer present was excluded, and the accused was acquitted. The confession was almost certainly true.

Cases of this shape are the hardest test of a rule, because the rule produces a visibly wrong outcome in the individual case in order to produce a better set of outcomes overall. The reasoning is not that the confession was unreliable, although confessions obtained under pressure often are. It is that the only sanction available against the practice is to make it useless. If evidence obtained improperly is admitted whenever it turns out to be accurate, the incentive to obtain it properly disappears, and the cases where it is inaccurate are precisely the ones nobody can identify from the inside.

The counter-argument is strong and usually underrated. The cost of exclusion falls on a victim in a particular case, while the benefit is spread across future cases and is invisible. Several jurisdictions have therefore narrowed the rule: evidence is excluded only where the breach was serious, or where admitting it would bring the administration of justice into disrepute, and a court weighs the two. Others keep a near-automatic rule on the ground that a balancing test is applied case by case by the institution whose behaviour the rule is meant to constrain.

Which design is better is an empirical question that is difficult to study, because the thing to be measured is the number of improper interrogations that did not happen.

What the debate makes clear is the shape of the choice. A rule that protects a process will sometimes produce an outcome that looks indefensible, and a system that abandons the rule whenever that happens does not have the rule at all.