The assumption that payoff-relevant information is observable but not verifiable is important for many core results in contract, organizational and institutional economics. However, subgame-perfect implementation (SPI) mechanisms - which are based on off-equilibrium arbitration clauses that impose fines for lying and the inappropriate use of arbitration - can be used to render payoff-relevant observable information verifiable. Thus, if SPI mechanisms work as predicted they undermine the foundations of important economic results based on the observable but non-verifiable assumption. Empirical evidence on the effectiveness of SPI mechanisms is, however, scarce. In this paper we show experimentally that SPI mechanisms have severe behavioral limitations. They induce retaliation against legitimate uses of arbitration and thus make the parties reluctant to trigger arbitration. The inconsistent use of arbitration eliminates the incentives to take first-best actions and leads to costly disagreements such that individuals - if given the choice - opt out of the mechanism in the majority of the cases. Incentive compatible redesigns of the mechanism solve some of these problems but generate new ones such that the overall performance of the redesigned mechanisms remains low. Our results indicate that there is little hope for SPI mechanisms to solve verifiability problems unless they are made retaliation-proof and, more generally, robust to other-regarding preferences.