Observed arrival · 2026-10-27
ProgramLab: teaching agents to rebuild real programs
A research dataset of 1,262 program-reconstruction environments for training and evaluating coding agents with reinforcement learning.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers studying agentic reinforcement learning and program synthesis
- Worth noticing
- The tasks span Go, Rust, and C/C++; their hidden test suites total 187,208 tests.
Field notes
Tasks are built from pinned repository commits and pair an execute-only reference with usage material such as help output, man pages, and fixtures. The page reports a median of 127 hidden tests per task and 76.6% median coverage of first-party lines; coverage exceeds 90% for projects under 5,000 lines. The task collection includes projects such as mkcert, jless, and Caddy.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue