Authorship verification of source code seeks to determine whether two programs were written by the same programmer, a task with applications in software forensics, plagiarism detection, and cybersecurity. Most recent deep learning approaches rely on embedding similarity and fixed decision thresholds, producing binary outcomes without probabilistic interpretation and offering limited support for combining evidence across multiple code fragments. This extended abstract presents ongoing work on ProbCLAVE, a probabilistic, set-based framework for source code authorship verification. Rather than proposing a new embedding model, ProbCLAVE reframes verification as probabilistic inference by calibrating similarity scores into posterior probabilities, transforming them into log-likelihood ratios, and aggregating evidence across multiple code samples per author. Preliminary results on a controlled Python dataset indicate substantial improvements in probability calibration, robustness to prior shifts, and verification accuracy compared to threshold-based baselines. We discuss the current status of the framework, initial empirical findings, limitations, and planned extensions toward multilingual, adversarial, and open-world scenarios.