Bonaldi et al. present and analyse the results of the Science Data Challenge 3a (SDC3a), a community-wide exercise on foreground removal for Epoch of Reionisation (EoR) experiments, organised by the Square Kilometre Array Observatory (SKAO). The challenge ran for eight months, from March to October 2023, and involved realistic SKA-Low data simulations spanning 106 MHz to 196 MHz. These simulations included foreground contamination from both extragalactic and Galactic emission, as well as instrumental and systematic effects. Participants were tasked with extracting cylindrical power spectra of the EoR signal, cleaned of all contaminations, and providing corresponding confidence levels.

Seventeen teams completed the challenge using a variety of methods, and their results were evaluated using multiple performance metrics. The outcomes suggest encouraging prospects for current foreground mitigation techniques, with seven teams achieving median errors in EoR power spectrum recovery that fall below the true signal level—though some notable outliers were observed. The smallest residual across all scales and frequencies was 4.2_−4.2⁺²⁰ × 10⁻⁴ K² h⁻³cMpc³. However, the accuracy of the confidence level estimations was generally lower, with most teams underestimating the true error, sometimes very significantly. The most accurate error bars captured about 60±20% of the actual errors. Overall, the challenge offered participants valuable insights into their pipeline performance, demonstrating that cross-comparison between independent methods could be a powerful strategy to identify biases and enhance error estimation in future EoR studies.

The figure shows the submissions from the top five teams in the SDC3a leaderboard, each represented by a single result, focusing on the frequency range 151–166 MHz. The left panel displays the power spectra submission P, which can be directly compared with the corresponding true power spectra P’ shown in Figure 2 in the paper. The centre panel presents the 1σ error bars ΔP as reported by each team, reflecting their estimated uncertainties. The right panel shows the true error committed, P − P′, representing the difference between each team’s submission and the true power spectrum. This side-by-side comparison highlights the strengths and limitations of each method in both signal recovery and uncertainty estimation.