We describe our system submitted to SemEval-2026 Task 5 on rating the plausibility of word senses in ambiguous sentences within narrative contexts. The task requires predicting human-perceived plausibility scores on a 1-5 scale for candidate word meanings embedded in short stories, posing challenges such as limited training data and the ordinal nature of target labels. Our approach combines a DeBERTa-v3-large encoder with Low-Rank Adaptation (LoRA) and a dynamically weighted hybrid CORAL-MSE loss for ordinal regression. This formulation adapts the contribution of ranking and regression objectives during training, prioritizing ordinal consistency early and regression refinement in later epochs.We analyze the contributions of dynamic loss weighting to overall system performance.