DESI DR2 BAO measurements prefer dynamical dark energy over a cosmological constant at 3.2-3.4 sigma. The standard CPL parametrization — w(a) = w₀ + wₐ(1-a) — captures this preference as a transition from phantom-like behavior (w < -1) at early times to quintessence-like behavior (w > -1) at late times. The transition crosses the phantom divide at w = -1, which has theoretical significance: it separates field theories with standard kinetic terms from those with exotic ones.
Bréval, Schimd, and Uzan (arXiv:2602.21169) ask what happens if you allow dark energy density to go negative. In the early universe, a negative dark energy density would be subdominant to matter and radiation — observationally invisible. But it changes the theoretical interpretation: the crossing of physical significance is not w = -1 but the null energy condition boundary (NECB), where ρ + p = 0. When density can change sign, these two boundaries are different.
The result: when negative-density phases are permitted, BAO and supernovae data push them beyond observational reach — the data doesn't constrain what it can't see. But the statistical preference for dynamical dark energy weakens. The 3.2 sigma becomes less significant because the model space is larger. The deviation from a cosmological constant that looked robust in the positive-density framework becomes less compelling when the framework is relaxed.
The general observation: the statistical significance of an anomaly depends on what you allow in the null hypothesis. Restricting the model space — here, requiring positive density — can make deviations appear more significant than they are. Expanding the space dilutes the preference. The strength of evidence depends not just on the data but on the size of the theoretical sandbox in which the data is interpreted.