NBA JAM Recomp prerelease brings rebuilt arcade games to Windows and Linux

A prerelease of NBA JAM Recomp, a rebuild of Midway's NBA JAM arcade games, is out for Windows 10/11 and Linux x64.

Arcade → PC (Windows), Linux / Steam Deck · Recompilation

olya_nro9 Nov 2026, 22:29· catalog:Decomp Buddy

NBA JAM Recomp v2026.11.06 is a prerelease app that plays Midway’s arcade basketball games on Windows 10/11 x64 and on Linux x64. The GitHub repository has no description yet, but the README explains the project.

The games are rebuilt from their original source code and played in one app with a game picker. Four titles are covered: NBA JAM Tournament Edition (rev 4.0) and NBA JAM T.E. Special Edition (rev 5.02) are built from source, with the Tournament Edition’s program and graphics matching the original ROM byte for byte. NBA Maximum Hangtime 1.3 is also built from source, with a matching program. The 1993 original NBA JAM and NBA Hangtime run straight from ROMs, because the original NBA JAM source was never published.

The repo ships no Midway code, graphics or sound. You copy your own files into a roms/ folder. The app accepts MAME romsets or the ZIPs of the public source repositories, scans sub-folders and ZIPs, and matches each file by CRC32. Built games go to data/<game>/ and are used when the ROMs are not found.

The app has CRT and sharp shaders, integer scaling, 4:3, 16:9, 16:10 and fill modes, and online play for up to 4 players. Netplay runs in lockstep over UDP and checks for ROM mismatches and desyncs. Keyboard and gamepad input both work.

How to try it:

  1. Download the Windows zip or the Linux tar.gz from the releases folder.
  2. Unpack it, create roms/ and copy in your MAME romsets or source ZIPs. The per-game list is in dist/NBAJamRecomp/README.txt.
  3. Run NBAJamRecomp.

The Linux build requires glibc 2.38 or newer. The README credits Pocket_lucho with building the project together with Claude, Anthropic’s AI model. That places it among AI-assisted projects in the scene. The README does not state how much of the code came from the model.

Sources: github.com