This service agreement concerns the Python FOSS library "PyMC", covering improvements to the reliability of the library’s operation, efficiency and speed. Achieving these goals entails designing and implementing better, tighter, handling of user configuration settings by means of validation and reporting deviations from expected patterns, as well as improving the documentation to inform users and downstream developers of the expected behaviour. It also entails designing tests for recently developed inference backends, in order to ensure conformance of these implementations with the expected behaviour across the supported backends, in addition to optimising and tuning performance requires the designing and running of benchmarks, and publishing a reproducible baseline report with conclusions. The work supports people learning PyMC from its example notebooks and downstream developers extending its log-probability machinery. Reliable configuration, visible progress, bounded compilation costs and executable examples help these groups obtain reproducible results without having to diagnose failures inside the underlying infrastructure. PyMC is shared statistical infrastructure rather than an end-user application. This work addresses researchers and practitioners who build statistical models, developers of scientific software that depends on PyMC, and contributors who maintain its compiler, sampler and documentation.