When a customer assigns three stars to a digital purchase, what are they actually saying? Rarely something simple. Analysis of 99,988 reviews across Amazon, Flipkart, and TripAdvisor reveals that 56% of three-star reviews simultaneously contain positive and negative signals about the same customer experience dimension — a structural complexity that star ratings, by design, cannot represent. Addressing the gap between CX theory and computational measurement, this study operationalizes the psychometrically validated TAPE construct — Trust, Autonomy, Personalization, and Engagement — as zero-shot natural language inference hypotheses, enabling annotation-free, dimension-specific CX measurement directly from review text. The framework requires no labelled data, no platform-specific retraining, and no domain adaptation — making it deployable wherever customers leave text feedback. By operationalizing these dimensions as natural language inference hypotheses applied to customer reviews, the framework achieves Pearson correlations of r = 0.658, r = 0.706, and r = 0.702 with star ratings across Amazon, Flipkart, and TripAdvisor respectively — outperforming VADER by 36%, TextBlob by 56%, and keyword lexicon approaches by 58% on Amazon, with consistent superiority maintained across all three platforms. Beyond aggregate performance, the framework reveals that customer experience has a contextually calibrated dimensional structure invisible to sentiment polarity tools: Trust dominates hospitality reviews, Personalization is most prominent in Indian e-commerce, and Autonomy — as an emerging dimension of digital commerce — shows the highest salience in US e-commerce contexts relative to other platforms, consistent with its theoretical characterization as a distinctively algorithmic-era CX concern. These platform signatures are undetectable by star ratings or general-purpose sentiment analysis. Outcome alignment analysis further confirms that Trust shows the strongest co-occurrence with satisfaction signals (r = 0.841, within-document estimate) while Engagement shows the strongest co-occurrence with loyalty signals (r = 0.639), providing dimension-specific intelligence for CX diagnosis.