Despite evidence that social interactions build social capital, evaluations show limited impacts of Community Driven Development (CDD) programs on social capital. We investigate whether this puzzle could result from an excessive reliance on surveys to measure social capital. Our study, in the context of a clustered randomized CDD in Nepal, initially indicates no impacts using surveys-based measures. However, incentivized lab-in-the-field games reveal increased cooperation, while network analysis unveils closer household connections. This highlights the sensitivity of program evaluations to social capital measurement choices and contributes to ongoing debates about empirically capturing social capital.