Tanveer Karim · interactive figure← Background & how to read this

Which data point is driving the case for evolving dark energy?

DESI DR2 BAO + Planck CamSpec + one supernova compilation, leave-one-tracer-out PSIS cross-validation. Every chain includes all three probes; the selector changes only which supernova compilation enters the joint fit. ΔELPD is written ΛCDM − w₀w_aCDM, so negative means w₀w_aCDM predicts that tracer better. The four lower panels show each model's predictive density for a withheld tracer on a shared axis measured in standard deviations of that tracer's own error, so the measurement always sits at 0 and the panels are comparable.

Supernova compilation

Predictive panels show

—

ΔELPD, 7 BAO tracers

—

joint, incl. SNe

—

Δp_loo (of 2 nominal)

—

withheld set

—

effect on the other probes & on w₀–w_a

Presets

Pareto k̂ k̂ ≤ 0.5 reliable 0.5 – 0.7 borderline k̂ > 0.7 unreliable |

ΛCDM w₀w_aCDM measurement ±1σ
Dark-energy panel: all data selected tracer withheld (68% credible, PSIS-reweighted). QSO (z=1.484) is omitted from the panel grid for space; it is well predicted by both models and is included in every number quoted above.