Decide whether DISCO-space or pixel-space likelihood is preferable for AMI image reconstruction

Ascertain whether regularised maximum-likelihood image reconstruction for JWST/NIRISS AMI data should be performed using a DISCO-space likelihood based on statistically independent kernel observables or directly in pixel space, and determine which choice yields superior sensitivity and fidelity.

Background

Amigo models finely sampled visibilities across the uv plane, enabling deconvolution and image reconstruction. The authors propose a JAX implementation akin to MPoL but note uncertainty about the optimal likelihood domain. Choosing between DISCO-space (statistically independent kernel amplitudes/phases) and direct pixel-space likelihoods has implications for computational efficiency, noise modeling, and reconstruction quality.