Ten verifiable environments, five scientific domains.
Every environment ships closed-form baselines, calibrated rewards, and frontier-model evals. Filter by domain to find your variant.
Single-turnSparse Fourier Recovery
Recover a k-sparse signal from m noisy Fourier measurements. Closed-form ground truth via OMP and L1.
Multi-turnSparse Fourier Recovery
Multi-turn variant: agent iteratively refines support estimates with feedback per round.
Tool-usingSparse Fourier Recovery
Tool-using variant: agent calls FFT, threshold, and least-squares primitives directly.
Single-turnCT Reconstruction (LoDoPaB)
Reconstruct low-dose CT slices from sparse-view sinograms. FBP and TV-regularized baselines.
Multi-turnCT Reconstruction (LoDoPaB)
Multi-turn LoDoPaB: agent iterates over filter / regularizer choices with PSNR feedback.
Single-turnMRI Knee (fastMRI)
Reconstruct knee MRI from undersampled k-space at 4× and 8× acceleration.
Multi-turnMRI Knee (fastMRI)
Multi-turn fastMRI: agent refines coil-combine and regularizer choices over rounds.
Single-turnPhase Retrieval
Recover phase from intensity-only measurements. HIO and Fienup-style baselines.
Multi-turnPhase Retrieval
Multi-turn phase retrieval: agent tunes support and shrinkage parameters across rounds.
Single-turnSuper-Resolution (DIV2K ×4)
4× upscaling of natural images. Bicubic, ESPCN, and SRCNN baselines for PSNR/SSIM scoring.