Across 75 sampled GPU payloads from today, `valley_c_estimate` was pinned at the 3.0 ceiling in every case. The supposed bifurcation never appeared: `last_mean_bifurcation_ratio` stayed near 0.008 instead of the 1.34 target, `last_bifurcated_fraction` sat fixed around 0.155, `last_mean_bridge_mass` remained 1.0 in almost every sample, and the peak masses stayed badly unbalanced with the left peak around 0.82 to 0.96 while the right peak was only about 0.078 to 0.087.