Data-Driven Variable-Exponent Analysis for Photoemission Yield Spectroscopy: An Autonomous Self-Diagnosing Framework Based on Integrated Residual Metrics

Photoemission yield spectroscopy (PYS) is widely used for evaluating the electronic states of materials. As automated materials discovery advances, unsupervised extraction of physical information from ambient-air PYS data becomes important. Conventional fixed-exponent analyses and logarithmic transformations suffer from heteroscedasticity, which destabilizes estimation in low-signal regions. To address this, we propose a data-driven analysis framework based on the 1/n-Scan method, which operates directly in the original signal space, removing the geometric bias inherent in log-transform approaches. We further integrate a self-diagnostic quality-evaluation system that quantifies estimation uncertainty with Akaike weights, together with independent residual metrics--the normalized mean absolute error (NMAE), the RMSE-to-MAE ratio (RMR), the Durbin-Watson (DW) statistic, and a macroscopic metric ($ΔR^2$)--that distinguish hardware-related data degradation from a physical-model mismatch. Applying the framework to differently doped Si and a polycrystalline Au reference in air, we demonstrate autonomous detection, without assumptions on the emission mechanism, of the breakdown of the single-component approximation in heavily doped p-type Si, arising from the overlap of two emission components with different thresholds, as a statistical anomaly--a decrease in DW below its critical value with an auxiliary increase in $ΔR^2$, despite sound NMAE and RMR--independently confirmed by a two-component fit ($Δ$AIC $\approx$ 57, DW recovering from 0.8 to 2.0). For heavily doped n-type Si, the gradual surface evolution was classified as a change within the single-component description. This framework provides a robust, self-diagnosing analysis engine for closed-loop autonomous materials exploration.