The Replication Renaissance: Why Science Starves for What It Won't Serve

The replication crisis is not a scandal of bad actors. It is a failure of physics. In any information-processing system, when the reward for signal novelty outpaces the structural penalty for noise, the system hallucinates.

Science is no exception. Recent text-mining of epidemiology and psychology literature (2024-2025) confirms a non-random spike in p-values between 0.045 and 0.049-the physical fingerprint of interpretive variance overwhelming experimental constraint. We formalize the crisis using the Coordination Invariant (S = C/V), derived from Shannon entropy: Scientific Stability = Constraint / Variance.

When V grows without proportional increase in C, mutual information between published claims and reality collapses exponentially. At current incentive ratios, replication project data converge on 70-80% of "significant" findings in high-impact journals being statistically indistinguishable from noise. The fix is not better scientists.

It is better physics. We propose three architectural interventions: (1) a Verification Score (Vs) weighting citations by independent replication status; (2) Registered Reports as default publication infrastructure; and (3) a Live Literature Protocol in which every paper carries a real-time stability designation. Registered Reports already publish null results at dramatically higher rates (54% vs.

21% in traditional workflows)-not because they fail, but because they finally report what is actually there. The Renaissance does not require moral reform. It requires constraint.

Build the C. Let the truth emerge.