Extended Data Fig. 5: TnC Selection Artificial Neural Network Definition.

The near detector (ND, left) and far detector (FD, right) Artificial Neural Network (ANN) cut definition are shown. Each plot shows full data (black-solid) and accidental background only (blue-solid) curves. The remaining data upon background subtraction is shown (black-points) represents correlated events, which are signal IBD-like. The 1σ uncertainty stands for 68% frequentist probability: statistics only (error bar). The IBD MC (solid red), with no backgrounds, is contrasted against the data. Sizeable differences between the FD and the ND ANN output are dominated by the different signal to background contamination of each detector. The ND has ~ 10 × better signal to accidental background. The FD has lower statistics. The MC exhibits excellent agreement to data across the entire dynamic for both detectors. A similar ANN definition had been demonstrated for FD-I data¹³. The ANN per detector cut was optimised to reduce the FD background and to match a slight prompt spectral distortion in both detectors (not shown explicitly). The latter is key to ensure an unbiased rate+shape θ₁₃ measurement. Such a distortion is known to arise from the Δr^{prompt–delay} variable slightly dependent on the prompt energy. Hence, the indicated ANN cut are slightly different for ND (0.86) and FD (0.85). This causes a 1.3% difference in rate normalisation, corroborated with data to a few per mille precision.