The fit audit — is your XPS fit defensible?

A flat residual is necessary but not sufficient: an XPS fit can match the data perfectly and still be chemically wrong, or use peaks the data doesn’t justify. PikaXPS’s fit auditor runs the checks an experienced reviewer would, in one click, and returns a ✖ / ⚠ / ✓ report.

What it checks

• FWHM sanity — flags widths that are unphysically narrow (noise-fitting) or too broad for a single species, and parameters stuck at their bounds.
• Reference binding energies — compares each peak’s position against the citation-backed database; if it matches no known state, it says so and names the nearest one.
• Doublet integrity — is the spin-orbit partner present, with the right splitting and area ratio? More on doublets →
• Lineshape — warns when a metallic state is fit with a symmetric Gaussian–Lorentzian (which inflates the oxide component).
• Expected satellites — flags missing shake-up satellites for states that always have them (Cu(II), Co(II), Ni(II)…).
• Over-fit signals — near-coincident peaks, <1% components, zero-area peaks.
• Charge referencing — is C 1s C–C on 284.8 eV?

The two statistical tests

• Residual z-score — band-pass filters the residual to peak-width features and reports the largest unexplained structure in units of the noise. Below 3σ, there’s nothing to add.
• Leave-one-out necessity (BIC) — refits the model without each peak and compares the Bayesian Information Criterion. If the fit is just as good without a peak, the data doesn’t require it. This is the test that catches the “I added a peak and nothing changed” situation — a strong sign the peak should go.

The rule

Statistics are necessary, not sufficient. Before adding a component, confirm a known chemical state can sit at that binding energy (the auditor checks the reference DB for you). Both the statistics and the chemistry have to agree.

Download PikaXPS → and run the audit on your next fit.