Extended Data Fig. 2. Time series to determine the experimental duration necessary to reach equilibrium.

N contents in the (a) alloy and (b) silicate melts, and consequently (c) $D_{\text{N}}^{\text{alloy/silicate}}$ show no variation beyond the uncertainties of the measurements as a function of time for experiments conducted at 3 GPa and 1600 °C for 0.5, 2, 6, and 12 hours (Experiment numbers: X63, G634, X74, and G639). These demonstrate that N exchange between the two phases had attained equilibrium at less than 0.5 hours at our experimental conditions. Also, an almost unchanged N content in alloy and silicate melts with increase in time means that there was no loss of N from the alloy + silicate melt system with increase in experimental run time. (d) NBO/T of the silicate melt compositions also show no variation with time beyond the uncertainties of the measurements, which illustrates that the silicate melt compositions had also reached steady state. All experiments were conducted with a fixed starting composition of alloy + silicate mixture (70%ThB1+30%Fe-5Ni-5N-17.5Si). Error bars in all panels are ±1-σ deviation based on replicate electron microprobe analyses and where absent the error bars are smaller than the symbol size.