Sentiment analysis remains the dominant computational approach for evaluating political news, yet its ability to capture the rhetorical complexity of political discourse is increasingly questioned. This paper presents a systematic comparison between a transformer-based sentiment classifier (RoBERTa) and a Large Language Model-based multi-dimensional framing analysis framework across 50 political news articles from 17 international outlets. While RoBERTa classifies 70% of articles as neutral and reduces political discourse to a three-way polarity scale, the LLM-based framework captures 13 numerical dimensions including bias direction and intensity, manipulation indicators (cherry-picking, loaded language, false equivalence), sensationalism, and communicative intent. Our correlation analysis reveals only weak-to-moderate relationships between sentiment polarity and framing dimensions (maximum Pearson r = 0.38, p < 0.01), demonstrating that these approaches measure fundamentally different properties of political text. Through case studies, we show that sentiment-neutral articles can exhibit extreme manipulation patterns, while highly negative articles may reflect factual reporting on inherently negative events. These findings argue for moving beyond sentiment as a proxy for media quality, toward multi-dimensional frameworks that can reveal the rhetorical strategies invisible to polarity-based analysis. All data and analysis code will be made available upon acceptance.

We propose a content-level measurement of cross-bloc framing divergence in news coverage of contested events, built on the eventstate (ρe) formalism of Bottino et al. (2026). For a given event, outlets are aggregated into editorially-coherent blocs and each bloc is represented by a density matrix ρ bloc on a 15-dimensional framing space. The trace distance D(ρ state, ρmainstream) measures cross-bloc divergence; benchmarking it against the withinWestern polarization D(ρ right, ρleft) controls for editorial variation. The top eigenvector of (ρ state − ρ mainstream) attributes divergence to specific framing axes. Two case studies (Hormuz blockade 2026, n = 16; Navalny death 2024, n = 14) demonstrate the construction. The work positions D(ρ state, ρmainstream) as a content-level observable of potential interest to threat-intelligence workflows that currently rely on source-level features.