Error-Correcting Output Codes (ECOC) decompose multi-class problems into a set of simpler subproblems, providing robustness through redundancy. In ordinal classification, however, standard ECOC designs may violate the natural order structure of the labels. In this work, we introduce an exhaustive ordinal ECOC framework that systematically generates all monotone partitions of the ordered class set. The proposed construction is obtained by considering every non-empty subset of admissible threshold positions, thereby deriving the complete family of contiguous ordinal groupings. The resulting coding matrix preserves ordinal consistency while increasing structural redundancy in a principled manner. Experimental evaluation on 46 benchmark ordinal datasets, using logistic regression as base learner and 30 independent runs, shows that the proposed method consistently outperforms both a nominal ECOC baseline and strong ordinal competitors. Statistical analysis confirms that the improvements are significant across datasets. These results demonstrate that exploiting the full space of monotone ordinal decompositions yields richer structural information than classical ordered partitions.