Bob writes in with a partial list of recent developments in the Stan community.

Governance: The interim Stan governing body stepped down and were replaced with a new board elected by the developer community.

Funding: Stan receives millions of dollars annually in grants, gifts, and in-kind contributions across its global developer base.

Releases: Stable quarterly releases across the toolchain continue with 2.21 being the latest.

Major new language and math library features: a new parser, abstract syntax tree, optimizer, and code generator written in OCaml with a slightly more flexible syntax paving the way for future developments, algebraic equation solvers, one-dimensional integrators, improved differential equation solvers, improved matrix derivatives and numerical stability, a thorough testing framework for first through third-order derivatives, model compile times reduced from 45s to 7s. [Actually, some models compile much faster than that! — AG] GPU and multi-core functionality has been extended and further optimized.

Methodology and tooling: we’ve developed improved convergence measures and better calibrated effective sample size estimates, improved simulation-based calibration measures, and included them in releases.

Interfaces: a new HTTP-server based PyStan, new command-line based wrappers CmdStanPy and CmdStanR, and Stan.jl (Julia) has been completely updated.

User community and teaching: community and teaching, there are fifty or more Stan short courses and tutorials around the world every year, several new books using Stan including one focusing on astrophysics, dozens of new case studies, dozens of short courses and hundreds of longer courses including several semester-length courses with video; there have been 1000 new forum registrations, 2500 forum threads, 17,000 forum posts, and 2 million forum page views.

Developer community: We’ve picked up a half dozen new regular developers with CmdStanPy, the math library, and the new parser being popular repositories. We’ve hired dedicated developers to work on dev ops and have improved our automatic testing from the math library through to end-to-end model efficiency. We’ve established guidance on proposing and reviewing new features, which has established designs for closures and ragged arrays in the language.

Bob emphasizes that these are just the items that came to mind; it’s not intended to be a complete list. Feel free in comments to add things that are not listed above.