We extract the top-quark mass value in the on-shell renormalization scheme from the comparison of theoretical predictions for pp → $$ t\overline{t} $$
t
t
¯
+ X at next-to-next-to-leading order (NNLO) QCD accuracy with experimental data collected by the ATLAS and CMS collaborations for absolute total, normalized single-differential and double-differential cross-sections during Run 1, Run 2 and the ongoing Run 3 at the Large Hadron Collider (LHC). For the theory computations of heavy-quark pair-production we use the MATRIX framework, interfaced to PineAPPL for the generation of grids of theory predictions, which can be efficiently used a-posteriori during the fit, performed within xFitter. We take several state-of-the-art parton distribution functions (PDFs) as input for the fit and evaluate their associated uncertainties, as well as the uncertainties arising from renormalization and factorization scale variation. Fit uncertainties related to the datasets are also part of the extracted uncertainty of the top-quark mass and turn out to be of similar size as the combined scale and PDF uncertainty. Fit results from different PDF sets agree among each other within 1σ uncertainty, whereas some datasets related to $$ t\overline{t} $$
t
t
¯
decay in different channels (dileptonic vs. semileptonic) point towards top-quark mass values in slight tension among each other, although still compatible within 2.5 σ accuracy. Our results are compatible with the PDG 2022 top-quark pole-mass value. Our work opens the road towards more complex simultaneous NNLO fits of PDFs, the strong coupling αs(MZ) and the top-quark mass, using the currently most precise experimental data on $$ t\overline{t} $$
t
t
¯
+ X total and multi-differential cross sections from the LHC.