Internet transport algorithms are foundational to the performance of network applications. But a number of practical challenges make it difficult to evaluate new ideas and algorithms in a reproducible manner. We present the Pantheon, a system that addresses this by serving as a community “training ground” for research on Internet transport protocols and congestion control (https://pantheon.stanford.edu). It allows network researchers to benefit from and contribute to a common set of benchmark algorithms, a shared evaluation platform, and a public archive of results. We present three results showing the Pantheon’s value as a research tool. First, we describe a measurement study from more than a year of data, indicating that congestion-control schemes vary dramatically in their relative performance as a function of path dynamics. Second, the Pantheon generates calibrated network emulators that capture the diverse performance of real Internet paths. These enable reproducible and rapid experiments that closely approximate real-world results. Finally, we describe the Pantheon’s contribution to developing new congestion-control schemes, two of which were published at USENIX NSDI 2018, as well as data-driven neural-network-based congestion-control schemes that can be trained to achieve good performance over the real Internet.