We present first self-consistent modelling of x-ray photoelectron spectroscopy (XPS) Ti 2p, N ls, 0 ls, and C ls core level spectra with a cross-peak quantitative agreement for a series of TiN thin films grown by dc magnetron sputtering and oxidized to different extent by varying the venting temperature Tv of the vacuum chamber before removing the deposited samples. So-obtained film series constitute a model case for XPS application studies, where certain degree of atmosphere exposure during sample transfer to the XPS instrument is unavoidable. The challenge is to extract information about surface chemistry without invoking destructive pre-cleaning with noble gas ions. All TiN surfaces are thus analyzed in the as-received state by XPS using monochromatic Al K alpha. radiation (hv = 1486.6 eV). Details of line shapes and relative peak areas obtained from deconvolution of the reference Ti 2p and N 1 s spectra representative of a native TiN surface serve as an input to model complex core level signals from air-exposed surfaces, where contributions from oxides and oxynitrides make the task very challenging considering the influence of the whole deposition process at hand. The essential part of the presented approach is that the deconvolution process is not only guided by the comparison to the reference binding energy values that often show large spread, but in order to increase reliability of the extracted chemical information the requirement for both qualitative and quantitative self-consistency between component peaks belonging to the same chemical species is imposed across all core-level spectra (including often neglected 0 is and C is signals). The relative ratios between contributions from different chemical species vary as a function of T-v presenting a self-consistency check for our model. We propose that the cross-peak self-consistency should be a prerequisite for reliable XPS peak modelling as it enhances credibility of obtained chemical information, while relying entirely on reference binding energy values introduces large ambiguity. (C) 2016 Elsevier B.V. All rights reserved.