Oblique Peer Review: Extending Pyragogy Design Patterns through Structural Isolation and the Limits of Blind Convergence

As human–AI co-learning evolves from the historical tradition of Peeragogy toward AI-native Pyragogy, the challenge is no longer merely collaboration but the deliberate introduction of structural friction capable of disrupting shared cognitive biases. This paper investigates how such friction can be engineered through the architecture of multi-agent critique systems. We formalize Oblique Peer Review (OPR), a 1+N architectural pattern composed of a single human orchestrator and N mutually blind AI agents operating without inter-agent communication. The central claim of OPR is structural rather than adversarial: critical tension emerges from the independence of analytical perspectives, not from opposition between conclusions. The agents do not debate, negotiate, or influence one another; they remain isolated throughout the evaluation process. Using Cognitive Rhythm Theory as a non-operationalized conceptual lens, we report a single-case deployment (n = 1, founder-as-researcher) of OPR within Obliqo, a production-grade critique engine. The deployment yielded an unexpected result. While OPR successfully prevented direct cross-contamination between agents, structural blindness alone did not guarantee analytical divergence. Under a macro-fault condition, multiple independent agents converged on the same underlying defect—an over-firing anomaly—despite complete isolation. More significantly, the study reveals a deeper epistemic limitation: agreement among independent, mutually blind models may itself constitute a form of spurious convergence. In the observed case, consensus emerged around a candidate mitigation that appeared robust precisely because it was independently reproduced, yet was later shown to be flawed when evaluated against recorded system data external to the review loop. These findings suggest that independence is a necessary but insufficient condition for epistemic robustness in multi-agent critique architectures. OPR therefore contributes not only a new Pyragogical design pattern but also an empirical caution: consensus among isolated evaluators should not be mistaken for validation. The paper concludes by examining the methodological limits of a single-case study, the reflexive circularity of using OPR to analyze OPR, and future directions for research on human–AI co-learning systems, structural critique, and the epistemology of distributed intelligence.