The 'Bidirectional Coherence Paradox' demonstrates that LLM performance and explanation quality can be inversely correlated depending on domain observability.
March 31, 2026
Original Paper
Coherent Without Grounding, Grounded Without Success: Observability and Epistemic Failure
arXiv · 2603.28371
The Takeaway
The paper proves that LLMs can act successfully while providing false mechanical explanations, or provide perfect explanations while failing to act. This directly challenges the common practitioner assumption that a model's ability to 'explain its work' is evidence of genuine grounding or understanding.
From the abstract
When an agent can articulate why something works, we typically take this as evidence of genuine understanding. This presupposes that effective action and correct explanation covary, and that coherent explanation reliably signals both. I argue that this assumption fails for contemporary Large Language Models (LLMs). I introduce what I call the Bidirectional Coherence Paradox: competence and grounding not only dissociate but invert across epistemic conditions. In low-observability domains, LLMs of